This study aimed to investigate the associations between activity behavior data and 2 proposed welfare categories (WC) based on productive performance, cyclicity, and absence of clinical disease. A single cohort of 202 Holstein dairy cows (Parity 1; n = 55; Parity ≥ 2; n = 147) was followed from 10 d pre-calving to 150 DIM. All cows were ear-tagged with a device measuring activity (AT), rumination (RT), and eating (ET) time (min/h). Daily milk yield and health data were retrieved from on-farm software. Ovarian cyclicity was evaluated by transrectal ultrasonography at 35 DIM. Upon completion of the follow up period, study cows were classified into one of 2 WC: (1) WC1 = milk yield ≥ study group mean, absence of clinical disease, and cyclic at d35; and (2) WC2 = cows that failed in at least one of the WC1 conditions. Differences in behavior data between WC were determined by repeated measures analysis, including, parity, DIM and the interaction term WC × DIM in the models. Subsequently, odds ratios were calculated using 5 min increments in behavioral variables to evaluate their ability on predicting WC categorization. Overall, 59 cows were classified in WC1 and 141 in WC2. A significant effect on AT was found for WC × DIM (P < 0.0001), but no differences were found in AT between WC. A significant effect on RT was determined for WC × DIM (P < 0.0001) and the effect of WC showed a tendency for significance (P = 0.08), where RT was greater in WC1 (20.0 ± 0.4 vs. 19.2 ± 0.23 min/h). Overall, ET did not differ between WC (WC1: 14.8 ± 0.5 vs. WC2: 13.7 ± 0.3 min/h; P = 0.25). As interaction terms were significant, multiple comparison analysis resulted in significant differences within the first 21 DIM, where WC1 had greater RT and ET. RT and ET were adequate predictors for WC; increases by 5 min in RT and ET increased by 2.2 (1.3 – 4.2; P < 0.001) and 1.6 (1.1 – 2.5; P = 0.02) times the odds of being classified as WC1. These associations could be used to implement welfare evaluations based on this behavioral parameters.