Three different bootstrap methods for estimating confidence intervals (CIs) for coefficient alpha were investigated. In addition, the bootstrap methods were compared with the most promising coefficient alpha CI estimation methods reported in the literature. The CI methods were assessed through a Monte Carlo simulation utilizing conditions comparable across previous research. Of particular interest was the impact of item nonnormality on the estimation methods. The results indicated a clear order in the performance of the estimation methods. The normal theory bootstrap method had the best performance by far as it had consistent acceptable coverage under all simulation conditions. If items were normally distributed or had small skewness, and computing power is an issue, the methods proposed by Bonett followed by the normal theory method were good alternatives. It should be noted that the Fisher method had high variability and unacceptably higher coverage across most of the investigated conditions. (Contains 5 tables and 2 figures.)