Objectives: Examine the probability of new active substances (NASs) approved in Canada between 1 January 1997 and 31 March 2012 acquiring a serious postmarket safety warning. Design: Cohort study. Data sources: Annual reports of the Therapeutic Products Directorate and the Biologic and Genetic Therapies Directorate; evaluations of therapeutic innovation from the Patented Medicine Prices Review Board and Prescrire International; MedEffect Canada website. Interventions: Postmarket regulatory safety warning or withdrawal from market due to safety reasons. Primary and secondary outcome measures: Compare the probability of acquiring a postmarket safety warning in Canada in four different groups of drugs: (1) traditional medications versus biologics; (2) medications that offer significant new therapeutic benefits versus those that do not. Determine how well the type of review that an NAS received from Health Canada predicted the product's postmarket therapeutic value. Results: The probability of a traditional NAS acquiring a serious safety warning and/or being withdrawn was 29.9% (95% CI 21.8% to 40.2%) vs 27.3% (95% CI 18.2% to 39.7%) for an NAS of biological origin (p=0.47, log-rank test). For medications that were significant therapeutic advances the probability was 40.2% (95% CI 24.5% to 60.9%) vs 33.9% (95% CI 26.4% to 42.7%) for those that were not (p=0.18, log-rank test). Health Canada was 77.4% accurate in predicting the therapeutic importance of an NAS. Conclusions: There was no difference in postmarket regulatory safety action between traditional medications and biologics and no difference between drugs with significant therapeutic benefits and those without. Although these results draw on Canadian data, they are likely to be relevant internationally. Further research should assess whether the current level of premarket safety evaluation is acceptable.