We compared reproductive performance across parity, lifetime performance and removal hazard of sows in commercial Japanese herds categorized into 3 sow reproductive performance groups and 2 herd productivity groups. The sow group categories were based on the lower and upper 25th percentiles of the number of pigs born alive (PBA) in parity 1, as follows: 8 pigs or fewer, 9–12 pigs and 13 pigs or more. The herds were classified into high- and low-performing herds on the basis of the 50th percentile of pigs weaned per mated female per year. We analyzed 213,514 parity records and 47,024 lifetime records of females entered into 96 herds. Multivariate and single response models were applied to reproductive performance across parity and lifetime performance, respectively, and a proportional hazard model was applied to removal hazard. Sows having 13 or more PBA in parity 1 had 1.0–1.4 more PBA in subsequent parities than sows having 8 or fewer PBA. However, there were no such differences between sow groups for weaning-to-first-mating interval across parity (P > 0.05). Also, sows having 13 or more PBA in parity 1 had 3.4–3.7 higher lifetime average PBA than sows having 8 or fewer PBA in both the herd groups (P < 0.05). Furthermore, the sows in the low-performing herds with 13 or more PBA in parity 1 also had 2.3 fewer lifetime average nonproductive days than sows having 8 or fewer PBA (P < 0.05), although again no similar difference was found for high-performing herds (P > 0.05). The removal hazards for a sow having 13 or more PBA in parity 1 were lower than those for a sow having 8 or fewer PBA (P < 0.05), with no difference in hazards between the 2 herd groups (P = 0.62). In conclusion, PBA in parity 1 can be used to predict a prolific or a low PBA sow. Also, producers in low-performing herds should pay particular attention to sows having low PBA in parity 1 to reduce nonproductive days.