To estimate sensitivity and accuracy of subjective evaluation of mild lameness in horses during treadmill locomotion and to correlate subjective evaluation with kinematic analysis. 19 lame and 5 clinically normal horses. Lameness was evaluated by subjective score and kinematic analysis before and after palmar digital nerve block (PDNB). Evaluations were made by 6 clinicians and 7 interns or residents. Within- and between-observer agreement analyses (kappa values) were calculated and compared, using a Student's t-test. Pearson's product-moment correlation coefficients were calculated between clinician's change in score and the change in kinematic variables after PDNB. Within-observer agreement was within the range expected for conditions of moderate diagnostic difficulty. Within-observer agreement was higher for clinicians than for interns or residents. Between-observer agreement was acceptable for scores within 1 value of each other. Between-observer agreement of change in lameness score after PDNB was poor. When kinematic variables were ranked with each clinician's subjective change in score, only 2 were among the top 3 for the majority of clinicians. Asymmetry of vertical head movement between contralateral forelimb stance phases and the point of maximum hoof height during swing decreased as lameness subjectively improved. Mild lameness may be difficult to evaluate during treadmill locomotion. Although clinicians were more repeatable in their subjective evaluation of lameness than interns or residents, they were not more reliable at detecting the true state of lameness. Lack of agreement between clinician scoring of mild lameness emphasizes the need to use more objective measures for quantifying lameness.