Most colorectal cancers are diagnosed after the onset of symptoms. However, the risk of colorectal cancer posed by particular symptoms is largely unknown, especially in unselected populations like primary care. This was a population-based case-control study in all 21 general practices in Exeter, Devon, UK, aiming to identify and quantify the prediagnostic features of colorectal cancer. In total, 349 patients with colorectal cancer, aged 40 years or more, and 1744 controls, matched by age, sex and general practice, were studied. The full medical record for 2 years before diagnosis was coded using the International Classification of Primary Care-2. We calculated odds ratios for variables independently associated with cancer, using multivariable conditional logistic regressions, and then calculated the positive predictive values of these variables, both individually and in combination. In total, 10 features were associated with colorectal cancer before diagnosis. The positive predictive values (95% confidence interval) of these were rectal bleeding 2.4% (1.9, 3.2); weight loss 1.2% (0.91, 1.6); abdominal pain 1.1% (0.86, 1.3); diarrhoea 0.94% (0.73, 1.1); constipation 0.42% (0.34, 0.52); abnormal rectal examination 4.0% (2.4, 7.4); abdominal tenderness 1.1% (0.77, 1.5); haemoglobin <10.0 g dl(-1) 2.3% (1.6, 3.1); positive faecal occult bloods 7.1% (5.1, 10); blood glucose >10 mmol l(-1) 0.78% (0.51, 1.1): all P < 0.001. Earlier diagnosis of colorectal cancer may be possible using the predictive values for single or multiple symptoms, physical signs or test results.