Abstract Background Previous case–control studies have reported positive associations of spicy food consumption with risks of certain gastrointestinal-tract (GI) cancers. However, there is no prospective evidence on such associations, particularly from China, where there are high incidence rates of GI cancers and spicy food is widely consumed. Methods The prospective China Kadoorie Biobank study recruited >512 000 adults aged 30–79 years from 10 areas in China during 2004–2008; 2350 oesophageal, 3350 stomach and 3061 colorectal incident cancer cases were recorded by 1 January 2017, after a median of 10.1 years of follow-up. Cox regression yielded adjusted hazard ratios (HRs) for each cancer associated with spicy food intake. Results Overall, 30% of participants reported daily spicy food consumption at baseline. Spicy food consumption was inversely associated with oesophageal cancer risk, with adjusted HRs of 1.00, 0.88, 0.76, 0.84 and 0.81 for those who never/rarely consumed (reference) and consumed monthly, 1–2 days/week, 3–5 days/week and 6–7 days/week, respectively (ptrend < 0.002). The association remained similar after excluding the first 3 years of follow-up but appeared stronger in participants who did not smoke or drink alcohol regularly (ptrend < 0.0001). The corresponding HRs for stomach cancer were 1.00, 0.97, 0.95, 0.92 and 0.89 (ptrend = 0.04), with the association disappearing after excluding the first 3 years of follow-up. For colorectal cancer, the HRs were 1.00, 1.00, 0.95, 0.87 and 0.90, respectively (ptrend = 0.04) and the inverse association appeared to be restricted to rectal rather than colon cancer (pheterogeneity = 0.004). The types and strength of spice used showed little additional effects on these associations. Conclusion In Chinese adults, higher spicy food consumption was associated with lower risks of certain GI cancers, particularly among individuals who never smoked or drank alcohol regularly.