Single-method assessment of physical activity (PA) has limitations. The utility and cross-validation of a composite PA score that includes reported and accelerometer-derived PA data has not been evaluated. Participants attending the Year 20 exam were randomly assigned to the derivation (two-thirds) or validation (one-third) data set. Principal components analysis was used to create a composite score reflecting Year 20 combined reported and accelerometer PA data. Generalized linear regression models were constructed to estimate the variability explained (R2) by each PA assessment strategy (self-report only, accelerometer only, composite score, or self-report plus accelerometer) with cardiovascular health indicators. This process was repeated in the validation set to determine cross-validation. At Year 20, 3549 participants (45.2 [3.6] y, 56.7% female, and 53.5% black) attended the clinic exam and 2540 agreed to wear the accelerometer. Higher R2 values were obtained when combined assessment strategies were used; however, the approach yielding the highest R2 value varied by cardiovascular health outcome. Findings from the cross-validation also supported internal study validity. Findings support continued refinement of methodological approaches to combine data from multiple sources to create a more robust estimate that reflects the complexities of PA behavior.