We began the modeling process with the idea that a fully cross-lagged panel analysis would provide valuable information on the relations between constructs in an inte- grated model of HIV risk. There has been limited guidance in the literature on how to work with models containing several constructs over multiple time points. Two sep- arate models (one with UP paths only and the other with DOWN paths only) were an- alyzed to obtain approximate start values for the larger models. The full model re- sulted in issues with suppression due, most likely, to collinearity among some same-construct measures. To overcome this, two separate models were designed in which either Time 1 behavior constructs predict cognition constructs or those same constructs predict Time 3 behavior. The results for this model were presented, and a final model-testing procedure was performed to reduce the model and thus to ascer- tain the “true” direction of causation between Decisional Balance and Self-Efficacy for Condom Use. Various aspects of this study are highlighted in the following.