This study investigated the levels, trends and determinants of contraceptive use-failure in Matlab, Bangladesh, using a set of prospective data on 25,960 women of reproductive age. The data were extracted from the Record Keeping System (RKS) of Matlab for the period 1978–94. If there was any live birth during the use or within 7 months after the discontinuation of use, it was considered as a failure. The life table technique and hazard model were used as analytical tools. The results suggest that use-failure for pills, IUDs (TCu 200) and injectables and other temporary methods increased from 1978 to 1988, but began to decline after 1988. The cumulative probability of first-method failure within 1 year of method acceptance of the cohort of 1990–94 acceptors was 12·9% for pills, 2·0% for IUDs, 0·5% for injectables, 22·0% for condoms and 13·4% for ‘other’ methods (sampoon, foam, jelly and traditional methods). For pills, condoms and ‘other’ methods, the likelihood of failure declined with the duration of use; by contrast, the probability of an IUD failure increased over time, peaking at 3 years of use. The injectables maintained a low likelihood of failure regardless of the duration of use. The quality of Community Health Workers' (CHWs) performance was associated with the risk of failure of all temporary methods except condoms; women's background characteristics associated with failure varied by method. The effect of the quality of the CHWs' performance and the background variables on failure did not change much over time. It is felt that contraceptive failure deserves the serious attention of programme managers and policy makers to make the Bangladesh national family planning programme more successful.