The Bayesian information criterion (BIC) can be useful for model selection within multilevel modeling studies. However, the formula for BIC requires a value for N, which is unclear in multilevel models, since N is observed in at least two levels. The present study uses simulated data to evaluate the rate of false positives and power when using a value for N of: level-1 N, effective N, and level-2 N under various sample size and ICC values. Result indicate that using level-2 N value for computation of BIC results in the highest power while still maintaining reasonably low false positive rates.