In education research, conceptual models are becoming more complex and more challenging to fit. In response, many researchers have turned to multi-step model fitting processes; first fitting confirmatory factor analyses (CFA) to their data, then extracting the latent factor scores; converting them to observed variables to use in additional analyses. In this simulation study, we investigate the role of sample size, factor loadings, and true between-factor correlation when examining whether factors extracted from CFAs should be treated as orthogonal or allowed to correlate. We found both the orthogonal and correlated methods to be biased, with the magnitude and direction dependent on the size of the true correlation and the extraction method used. Results are discussed regarding impact on current practice.