To develop better ways to use behavioral data to detect lame cows, comparisons of milking visit behavior, milk yield, and lying behavior were made between lame and sound cows on automated milking system (AMS) farms. Data were collected for 30 cows, over a 6-d period, from each of 26 AMS farms in Eastern Ontario, Canada. Cows were gait scored using a 5-point numerical rating system (1 = sound to 5 = lame). Cows with gait scores < 3 were classified as sound (n = 527) while those with gait scores ≥ 3 were classified as lame (n = 245). Milking visit behavior was extracted from the AMS computer at each farm, while lying behavior was measured continuously using electronic data loggers. Body condition and hygiene were also scored, and were compiled with other cow-level variables such as parity and DIM. Data were summarized across the 6-d observation period/cow and analyzed in multivariable general linear mixed models. When controlling for DIM, milking frequency (#/d; mean ± SE) was lower (P < 0.001) for lame cows (2.8 ± 0.1) compared with sound cows (3.1 ± 0.1). Milk yield was lower (P = 0.05) for lame cows (32.0 ± 0.9 kg/d) compared with sound cows (33.4 ± 0.8 kg/d), while accounting for parity, DIM, and body condition. Compared with sound cows, lame cows had fewer AMS refusals/d (1.0 vs. 1.8; SE = 0.2; P < 0.001). Lameness status did not affect the frequency of AMS failures/d (0.1 ± 0.01). Overall, lame cows made fewer visits to the AMS than sound cows (4.0 vs. 5.0 visits/d; SE = 0.3; P < 0.001). When accounting for parity and DIM, lying time was greater (P = 0.002) for lame cows (728.2 ± 13.7 min/d) compared with sound cows (693.8 ± 12.0 min/d). These results demonstrate the differences in behavior and resulting productivity between lame and sound cows, indicating the potential use of behavioral indicators to identify lame cows on AMS farms.