Abstract Mammographic density (MD) is an established predictor of breast cancer risk. However, there is limited information on the robustness of the risk associations for different study designs and the associated methodologic challenges. Our analysis includes 165 samples from studies published since 2006. We use a weakly informative Bayesian approach to avoid unduly optimistic estimates of uncertainty, as found in the previous literature. We find that the existing consensus from previous review studies has underestimated the strength and precision of MD as a risk marker. Moreover, although much of the published literature is based on categorical measurement of MD, there are tangible advantages in using continuous data in terms of estimate precision and relevance for different patient populations. Estimates based on the percentage of MD are more precise for lower density women, whereas absolute MD has advantages for higher density. We show that older results might not be a good proxy for current and future findings, and it would be pertinent to adjust clinical interpretations based on the older data. Using an appropriate estimation method cognizant of the importance of heterogeneity is critical to obtaining reliable and robust clinical findings that are relevant for broad patient populations.