Research dossier 三线建设 · the Third Front, 1964–1980

The Third Front

From 1964 China moved its war industry into the mountains of the interior, explicitly because of the atomic bomb. It consumed two fifths of national construction investment for sixteen years, and it is the reason a target list drawn from population centres misreads the country.

Compiled 11 September 2026 The distances in Section 04 were measured against this engine’s own target list

Why this is here

This project chooses targets, and where no target list is on the record it chooses them the way every open assessment does: by population, largest cells first. That rule is stated on every readout and it is defensible, because a country’s industry and its people are mostly in the same places.

Mostly. The Third Front is the largest exception any state has ever built — a deliberate, sixteen-year, nationally impoverishing effort to put the things worth destroying where the people are not. This dossier exists because a model should know where its own method fails, and this is where it fails against China.

It is also the answer to a question the 1969 dossier raises and does not settle: what a strike on the Chinese nuclear complex would have achieved. Part of the answer is that by 1969 the complex was already moving.

Evidence tiers used throughout

DocumentedStated in a named source at a locatable place, or computed here from data this project ships.

InferredA reading of published figures, or a composite assembled from more than one source.

ContestedPublished assessments disagree, or the figure rests on official data whose basis is not published.

Not establishedWidely stated and not traceable, or never disclosed at all.

Filter claims
SECTION 01

What it was

A rear area, built from nothing, in the hardest country China had.

DocumentedThe country was divided into three fronts, and the industry was moved to the third. The first front was the coast, which an invader reaches first. The second was the middle. The third was the mountainous interior — Sichuan, Guizhou, Yunnan, the west of Shaanxi and Gansu, Qinghai and Ningxia, and the parts of Hebei, Henan, Hubei and Hunan west of the Beijing–Guangzhou railway. A Small Third Front was run by the provinces in their own mountainous interiors, the Shanghai programme being the largest of those.

DocumentedMao’s own reason was the bomb, and he gave it in those words. “In the period of the atomic bomb, we need a strategic rear for retreat.” The programme was not a response to the American war in Vietnam, whatever the sequence suggests: on 10 May 1964 Mao rejected the draft Third Five-Year Plan’s agriculture-first emphasis and demanded the Panzhihua steel base be accelerated, and on 27 May he convened an enlarged Politburo Standing Committee meeting on the Third Front. The Gulf of Tonkin incident was 2 August — three months later. It catalysed the mobilisation; it did not cause the decision.

DocumentedThe rule for siting was three words, and they are the whole design. “Near the mountains, dispersed, in caves” — kaoshan, fensan, jindong. Plants were put in valleys where a ridge stood between them and anything else, split into pieces far enough apart that one weapon could not take two, and where the geology allowed, put underground. Plant 816 at Fuling, an entire plutonium reactor complex dug into a mountain, is the extreme case, and it was never finished.

InferredThe design is legible as counter-targeting because it answers each part of the problem in turn. Dispersal defeats area coverage: a laydown sized to a city cannot reach plants seventy kilometres apart. Terrain defeats the outer rings, since a ridge takes the overpressure down. Depth defeats everything but a ground burst directly overhead. Nobody in Beijing published a targeting analysis, so this reading is inference — but it is the reading the geography forces.

SECTION 02

The scale

Larger than any other thing the Mao era did, and unacknowledged for fifteen years after it stopped.

The Third Front, 1964–1980
MeasureFigureSource and note
Share of national basic construction investment39.01%205.26 billion yuan. Chinese official-derived. Note the units error in wide circulation: 2,052.6 亿元 is 205.26 billion yuan, not 20.52 billion
Share of state investment to armaments at the peak11%1969 to 1971 — Naughton, The China Quarterly 115 (1988). This is the quantitative signature of the Soviet shock, and it is the clearest thing in the record
Sichuan industrial investment to military industry, 1950–8122.8%Naughton (1988)
Large and medium projects1,100+1,945 industrial enterprises and research institutions in all; about 400 state enterprises relocated inland
Workers moved inland, 1964–70145,000With 38,000 units of equipment, across 380 large projects
People mobilised in all~15 millionMeyskens, Mao’s Third Front (Cambridge University Press, 2020)

DocumentedWhat it cost, in the judgement of the economist who first measured it. Barry Naughton: the programme “was immensely costly, having a negative impact on China’s economic development that was certainly more far-reaching than the disruption of the Cultural Revolution”. Covell Meyskens: it received “more government investment than any other developmental initiative of the Mao era”, and was not officially acknowledged for over a decade and a half after it ended.

Documented1969 is visible in the figures. The border war at Zhenbao triggered a second investment surge, generally dated 1969 or 1970 to 1971, and turned the strategic emphasis north and north-west — against the Soviet Union — where it had been south-west, against the United States. The nuclear complex moved with it: construction at the Guangyuan plutonium plant began in October 1969, during the crisis, and the Ninth Academy’s relocation from Qinghai to Mianyang began in the same month.

Not establishedWhat it actually built, as against what it cost. No completion rate, no output series and no plant-by-plant accounting is published. The figures above are inputs. A programme can absorb two fifths of a country’s construction investment and produce very little, and several of its largest projects — Plant 816 among them — produced nothing at all.

SECTION 03

The nuclear complex, before and after

The clearest case, because both halves of it are documented.

Where the fissile-material complex was, and where it was going
FacilityFirst front, as builtThird Front replacementBegun
EnrichmentPlant 504, in Lanzhou, beside the Yellow RiverPlant 814 at Heping, on the Dadu in Sichuan1966
PlutoniumPlant 404, on the Gobi near YumenPlant 821 at Guangyuan, SichuanOct 1969
Plutonium, undergroundPlant 816 at Fuling, inside a mountainFeb 1967
Weapons design and assemblyPlant 221, on the Qinghai grasslandMianyang, SichuanOct 1969
Missile test rangeBase 20, JiuquanBase 25 at Kelan, ShanxiMar 1966
Fuel fabricationPlant 202, in BaotouPlant 812 at Yibin, Sichuanlate 1960s

DocumentedEvery node of the complex had an interior replacement under construction, and in 1969 not one of them was working. Plant 814 opened in June 1970, nine months after the crisis. Plant 821’s reactor went critical in December 1973. Plant 405 at Hanzhong, decided in 1969, did not operate until the 1990s. Plant 816 was begun in February 1967 and never completed at all.

Which cuts both ways, and the dossier on 1969 should say so

A Soviet strike in 1969 would have caught the whole of the operating complex on the first front, because the second one did not exist yet. That is the case for the strike having been feasible, and it is why the arithmetic in the nuclear surgery comes out as cleanly as it does.

But it would not have ended the programme, and the reason is in the same table. The replacements were dug, funded and under way; the weapons already built were not at these addresses; and the response to being struck would have been to accelerate what was already the largest construction programme in the country. The strike destroys a complex. It does not destroy the decision to have one, and by 1969 that decision had five years of concrete behind it.

SECTION 04 — CORE

What this does to a target list

Measured against this project’s own rule, on this project’s own data.

DocumentedThe rule, stated as the engine states it. Where no target list is on the record, this project takes the most populous five-arc-minute cells of a gridded census, largest first, skipping any cell within forty kilometres of one already taken. It is the rule the open literature uses and the rule the nuclear-winter papers use to set their fuel loadings. Every readout that uses it says so.

DocumentedApplied to China, it misses the Third Front by between seventy and two hundred and fifty kilometres. The distances below are computed here, from the engine’s own Chinese target list — the same file the winter model marks its fires with — to eight Third Front sites whose coordinates are in the record.

Distance from each site to the nearest struck cell, by the size of the list
Third Front siteNearest of the top 100Nearest of the top 250
Plant 821 · Guangyuan245 km154 km
Panzhihua steel base198 km4 km
Plant 814 · Jinkouhe168 km160 km
Plant 405 · Hanzhong145 km145 km
Base 25 · Kelan125 km125 km
Ninth Academy · Mianyang104 km3 km
Plant 816 · Fuling84 km42 km
Plant 812 · Yibin71 km6 km

Seventy kilometres is not a near miss

A hundred-kilotonne air burst puts its outermost ring — one pound a square inch, where windows break — about ten kilometres out. A megatonne reaches twenty-two. The closest of these eight sites to a hundred-weapon countervalue laydown is seven times beyond the outer ring of the weapon aimed at the nearest city, and the furthest is more than ten times beyond it. Nothing on that list touches any of them.

Widening the list to two hundred and fifty helps in three cases and not in the other five, and it helps for a reason that is not reassuring: Panzhihua, Yibin and Mianyang came within a few kilometres because towns grew around the plants. Sixty years of a works in a valley produces a population cell. The method finds the Third Front only where the Third Front stopped being hidden.

The three that stay out of reach at two hundred and fifty cells — Guangyuan, Jinkouhe, Hanzhong, Kelan — are the ones that stayed small. Those are also, in two cases, the plutonium and the enrichment.

InferredWhat follows for a model, and it is not that the model is wrong. A population rule answers a countervalue question correctly: if the object is the population and the urban-industrial base, the largest cells are the largest cells. What it cannot do is answer a counterforce question against a state that has spent two fifths of its construction budget on making the two lists different. Against the Soviet Union, or the United States, or Britain, the two lists overlap heavily. Against China after 1964 they do not, and a study of a strike on China that draws its targets by population is answering a question about cities while appearing to answer one about war-making capacity.

Not establishedWhat a real counterforce list against China would contain. No such list is public from any state. This project therefore does not have one, and where it strikes Chinese targets it strikes either the documented nuclear complex — as the 1969 study does — or population cells with the rule named on the readout. It does not have a third option, and inventing one would be worse than saying so.

SECTION 05

Did it work?

Unanswerable in the way that matters, and answerable in two ways that do not.

Not establishedIt was never tested, so the military question has no answer. No weapon was ever fired at the Third Front. Whether dispersal into mountain valleys would have preserved a war economy under attack is a counterfactual about a war that did not happen, and no published analysis attempts it. Anybody who tells you it would or would not have worked is reasoning from the design rather than from evidence.

DocumentedAs economics it was a heavy net loss, and China itself concluded so. From 1980 the programme was wound down, and from 1983 a formal adjustment moved or closed the least viable plants — the ones put in valleys so remote that supplying them cost more than they produced. Naughton’s verdict stands as the field’s: more far-reaching in its damage than the Cultural Revolution.

ContestedAs development policy the reckoning is genuinely open. The revisionist case is that the interior provinces got industry, railways and technical education they would not otherwise have had for decades, and that some of the present industrial base of Sichuan and Guizhou traces to it. The critical case is that the same investment anywhere else would have produced more of everything. Both are argued from the same figures, because the counterfactual is unobservable.

InferredThe one thing it certainly did is the thing this dossier is about. Whatever it cost and whatever it built, it made the country harder to target, and it did so at a scale no other state has attempted. That is a fact about geography rather than about economics, and it survives every verdict on the rest of the programme.

SECTION 06

What is not established

Including two things this dossier would need in order to be more than it is.

Not found, or never disclosed

  • A plant-by-plant list with coordinates. The eight sites measured in Section 04 are the ones whose positions are in the open record because they are nuclear. The other nineteen hundred are not mapped anywhere public, so the distances above are a sample and not a survey.
  • Completion and output figures. The published numbers are all inputs — investment, projects begun, people moved. What was finished and what it produced is not published.
  • Any Chinese targeting analysis justifying the dispersal distances. The siting rule is documented; the reasoning that chose seventy kilometres rather than thirty is not.
  • The cost of the shelter programme as distinct from the Third Front, and any credible national total for shelter area. The often-quoted claim that a city’s tunnels had “capacity for all six million inhabitants” is a government statement, not an engineering assessment.
  • Naughton (1988 and 2024) and Meyskens (2020) in full. The figures here come from abstracts, indexed extracts and secondary summaries, and are unverified against the works themselves. That is a real limitation of this dossier and it is stated rather than hidden.
SOURCES

Principal references