The reproducibility of ballistic actions performed on a custom-made apparatus was assessed in six untrained men who reported to the laboratory two times over a span of 10 d, on each occasion performing three isometric maximal voluntary contractions (MVC, elbow extension) and 10 ballistic elbow extension movements in each of two conditions: unloaded (L0) and with a load equal to 10% MVC (L10). For both the average of the trials and the best trial, method errors and Pearson correlation coefficients (r) were calculated for isometric peak torque and 10 (5 at each of L0 and L10) selected ballistic movement parameters (peak torque, peak rate of torque development, movement time, peak velocity, and peak acceleration). Based on method errors obtained and given an alpha level of 0.05 and power of 0.9, needed sample sizes were calculated for treatment effects ranging from 5 to 20%. Method errors for isometric peak torque were 5.1 and 3.1% for best and average of trials, respectively. For the 10 ballistic parameters, the corresponding method errors averaged 7.1 +/- 5.4% (SD) and 6.8 +/- 3.6%. The largest method errors were found in peak torque and peak rate of torque development with L0. The expected high negative correlation between method errors and r values was not found. For 5, 10, and 20% treatment effects, 4, 6, and 9 of 10 best trial measures met the sample size criterion of N < or = 20. For the average of trials, 3, 8, and all 10 met the criterion. With the exception of rate of torque development, the reproducibility of the described ballistic measures is acceptable for longitudinal training studies and cross-sectional group comparisons.