OBJECTIVES: To examine item response patterns over time using the SF-12v2™ from a measurement perspective using US national representative samples. METHODS: Four panel data with two-year repeated measures on each respondent were extracted from the Medical Expenditure Panel Survey (MEPS). Respondents were included if they were ≥ 18 years, had completed SF-12v2™ and, had at least one of the top ten most prevalent health conditions identified using ICD-9-CM. Three-facet measurement model was used to parameterize time as a distinct facet in the model, in addition to person and item facets. Interactions between time and the twelve items were examined at each time point in all panels. Goodness-of-fit of the items to the model was examined in repeated measures as well as in point-in-time measures. INFIT mean-square (MnSq≤1.40) was used as an item fit indicator. Cross-validations were conducted in each disease groups. RESULTS: Four panels were comparable in their distributions in health conditions, socio-demographics (mean ages were 52-53 years and, about 76-77% were white) and, sample sizes (2003-04, n=2,124; 2004-05, n=2,070; 2005-06, n=2,148 and 2006-07, n=2,329). Consistently in all panels, significant time and item interaction biases were found at time 1, especially on mental health items (p<0.01). On the other hand, interaction biases between time and items at time 2 were not significant (p>0.05). All items fit the model in repeated measures where time was parameterized as a facet (INFIT MnSq≤1.40). The mental health item “Have you felt calm and peaceful?” consistently showed misfit in all point-in-time measures (INFIT MnSq>1.40). Similar findings were noted in sub-samples.  CONCLUSIONS: Findings from this study suggest consistent learned response patterns over time, especially the responses to mental health item, which give rise to the importance of inter-temporal health context in health measurement. Hence, cross-sectional health measures should be interpreted with caution.