Background: Mammographic density (MD), the area of the mammographic image that appears white or bright, predicts breast cancer risk. We estimated the proportions of variance explained by questionnaire-measured breast cancer risk factors and by unmeasured residual familial factors. Methods: For 544 MZ and 339 DZ twin pairs, and 1,558 non-twin sisters from 1,564 families, MD was measured using the computer-assisted method Cumulus. We estimated associations using multi-level mixed-effects linear regression and studied familial aspects using a multivariate normal model. Results: The proportions of variance explained by age, body mass index (BMI), and other risk factors, respectively, were: 4%, 1% and 4% for dense area; 7%, 14% and 4% for percent dense area; and 7%, 40% and 1% for non-dense area. Associations with dense area and percent dense area were in opposite directions than for non-dense area. After adjusting for measured factors, the correlations of dense area with percent dense area and non-dense area were 0.84 and -0.46, respectively. The MZ, DZ and sister pair correlations were: 0.59, 0.28 and 0.29 for dense area; 0.57, 0.30 and 0.28 for percent dense area; and 0.56, 0.27 and 0.28 for non-dense area (standard error (SE) = 0.02, 0.04 and 0.03, respectively). Conclusions: Under the classic twin model, 50-60% (SE = 5%) of the variance of MD measures that predict breast cancer risk are due to undiscovered genetic factors, and the remainder to as yet unknown individual-specific, non-genetic factors. Impact: Much remains to be learnt about the genetic and environmental determinants of MD.