States and traits are important concepts in psychological research. They can be operationalized (a) by using measures that employ different time frames and (b) by applying statistical models that decompose the variance. However, the effects of using variations in states and traits by applying measurement and modeling approaches have yet to be merged and studied systematically. The present study addressed this topic by revisiting an intensively studied research question, namely: What is the longitudinal relation between self-esteem and depressive symptoms? To do so, we created state and trait versions of questionnaires by systematically changing the time frame ("during the last 2 weeks" vs. "in general") that was used to measure self-esteem and depressive symptoms and in addition, by using state-trait statistical models. We conducted an exploratory study (N ϭ 683) and a confirmatory replication study (N ϭ 1,087) with samples of university students, designed as a 2 ϫ 2 longitudinal experiment with 4 time points spanning 1 semester. Our results indicated that first, consistently across the 2 studies, trait time frames revealed higher proportions of trait variance than state time frames. Second, across the 2 studies, the well-researched vulnerability effect, which postulates that low self-esteem predicts depressive symptoms, only held when trait time frames for self-esteem were applied and traditional cross-lagged models were used. Third, when controlling for stable trait differences, cross-lagged results were least consistent when trait time frames were used, which highlighted the interdependency involved in measuring and modeling states and traits.