Prior literature has documented improvements in cancer survival over time. However, ambiguity remains over the relative contributions of improved treatment and earlier detection to survival gains. Using registry data, we developed a novel framework to estimate the relative contributions of advances in treatment and detection. Our approach compares changes in the probability of early detection, which we interpret as the effects of advances in detection, to improvements in stage-conditional survival, which we interpret as the effects of treatment. We applied this methodology using SEER data to estimate probabilities of early detection and stage-conditional survival curves for several cancers, by race, between 1988 and 2000. Survival increased for all of the cancers we examined, with blacks experiencing larger survival gains than whites for all cancers combined. Our baseline analysis found that treatment advances account for the vast majority of survival gains for all the cancers examined: breast cancer (83%), lung cancer (85%), colorectal cancer (76%), pancreatic cancer (100%), and non-Hodgkins lymphoma (96%). Compared to whites, treatments appear to explain a lower percentage of survival gains for blacks for all cancers combined; breast cancer, NHL, and pancreatic cancer show a higher percentage of survival gains than lung cancer; and roughly the same percentage for the colorectal cancer. These results are robust to sensitivity analyses examining potential length and lead time bias. Overall, our results suggest that while improved treatment and early detection both contributed to the recent gains in survival, the majority of gains from 1988 to 2000 appear to have been driven by better treatment, manifested by improved stage-conditional survival. These results have important policy implications regarding investment in research and development and the evaluation of efforts to improve cancer screening.