The work of R. MacCallum et al. (1992) was extended by examining chance modifications through a Monte Carlo simulation. The stability of post hoc model modifications was examined under varying sample size, model complexity, and severity of misspecification using 2- and 4-factor oblique confirmatory factor analysis (CFA) models with four and eight secondary loadings respectively and four levels of misspecification. Sample sizes were 200, 400, 800, and 1,200. Results in terms of recovery of population models partially support findings by MacCallum that more severe levels of misspecification result in less successful searches. The study also supports the sample size effect found by MacCallum, in which larger samples tend to result in greater recovery of the population models. Some evidence was found of differences due to model complexity, with model modifications somewhat more stable in the 4-factor model. An alternative to the number of modifications or overall nonsignificance as stopping criteria might be inspection of the modification indexes for a distinctive drop in successive values of the maximum modification index. Three tables and two figures present analysis results. (Contains 30 references.) (SLD)