Monte Carlo studies explored the sampling characteristics of Cohen's d and three approximations to Cohen's d when used as average effect size measures in meta-analysis. Reviews of 10, 100, and 500 studies (M) were simulated, with degrees of freedom (df) varied in seven steps from 8 to 58. In a two independent groups design, samples were obtained from populations whose mean differences represented a zero, small, medium, or large effect size. One thousand replications of studies within each of the 84 combinations of effect size, df, and M were conducted, and a mean and standard error were obtained for each combination of conditions. As expected, d was a positively biased estimator of effect size, overestimating by as much as 13 percent even with Hedges' correction factor. Surprisingly, the most unbiased estimator of effect size and the highest relative efficiency was obtained with the approximation to d computed from the obtained t and corrected according to Hedges. The approximations to d from the nonparametric statistic and the obtained conventional significance levels were not consistent estimators of effect size. These simulation results suggest that additional study of the behavior of effect size estimators should precede a more widespread application of meta-analysis. (Author)