Summary Background Four studies published since December, 1995, reported that the incidence of venous thromboembolism (VTE) was higher in women who used oral contraceptives (OCs) containing the third-generation progestagens gestodene or desogestrel than in users of OCs containing second-generation progestagens. However, confounding and bias in the design of these studies may have affected the findings. The aim of our study was to re-examine the association between risk of VTE and OC use with a different study design and analysis to avoid some of the bias and confounding of the earlier studies. Methods We used computer records of patients from 143 general practices in the UK. The study was based on the medical records of about 540 000 women born between 1941 and 1981. All women who had a recorded diagnosis of deep-vein thrombosis, venous thrombosis not otherwise specified, or pulmonary embolus during the study period, and who had been treated with an anticoagulant were identified as potential cases of VTE. We did a cohort analysis to estimate and compare incidence of VTE in users of the main OC preparations, and a nested case-control study to calculate the odds ratios of VTE associated with use of different types of OC, after adjustment for potential confounding factors. In the case-control study, we matched cases to controls by exact year of birth, practice, and current use of OCs. We used a multiple logistic regression model that included body-mass index, number of cycles, change in type of OC prescribed within 3 months of the event, previous pregnancy, and concurrent disease. Findings 85 women met the inclusion criteria for VTE, two of whom were users of progestagen-only OCs. Of the 83 cases of VTE associated with use of combined OCs, 43 were recorded as deep-vein thrombosis, 35 as pulmonary thrombosis, and five as venous thrombosis not otherwise specified. The crude rate of VTE per 10 000 woman-years was 4·10 in current users of any OC, 3·10 in users of second-generation OCs, and 4·96 in users of third-generation preparations. After adjustment for age, the rate ratio of VTE in users of third-generation relative to second-generation OCs was 1·68 (95% CI 1·04–2·75). Logistic regression showed no significant difference in the risk of VTE between users of third-generation and second-generation OCs. Among users of third-generation progestagens, the risk of VTE was higher in users of desogestrel with 20 g ethinyloestradiol than in users of gestodene or desogestrel with 30 g ethinyloestradiol. With all second-generation OCs as the reference, the odds ratios for VTE were 3·49 (1·21–10·12) for desogestrel plus 20 g ethinyloestradiol and 1·18 (0·66–2·17) for the other third-generation progestagens. Interpretation The previously reported increase in odds ratio associated with third-generation OCs when compared with second-generation products is likely to have been the result of residual confounding by age. The increased odds ratio associated with products containing 20 μg ethinyloestradiol and desogestrel compared with the 30 μg product is biologically implausible, and is likely to be the result of preferential prescribing and, thus, confounding.