Despite its popularity, the MEDmetric KT-1000 arthrometer's reliability remains inadequately documented. We conducted this study to determine the magnitude of trial-to-trial (within installation), installation-to-installation (within day), and day-to-day (between day) variability of anterior/posterior translation measurements in normal knees. We selected six normal subjects, three males and three females, and tested each on 6 consecutive days with three separate installations per day. We recorded the total anterior/posterior translation at +/- 89 and +/- 134 N force at 25 degrees of flexion during three consecutive trials in a single installation. Analysis of variance showed that no significant difference existed between trials (within installation) or between installations (within day) for all parameters. However, we did find a significant difference between days for individual right and left knee translation measurements at 89 and 134 N force. More importantly, no significant difference existed between days for right to left differences at both force levels. The magnitude of the expected measurement variability was expressed by computing 90% confidence limits for total anterior/posterior translation at +/- 89 N force. These were +/- 1.5 mm for the right knees, +/- 1.4 mm for the left knees, and +/- 1.6 mm for the right-left differences. Fischer's protected least significant difference post hoc test revealed that for all parameters, the 1st day measurements were significantly less than those on following days, suggesting that patient and examiner adjust to the testing procedure. We conclude that the standard KT-1000 evaluation should report paired differences rather than individual knee measurements. Additionally, initial evaluation should be supplemented by follow-up examinations for verifying translation values.