Faking remains an unsolved problem in high-stakes personality assessment. It is important that the evaluation of so-called faking-detection scales differs between psychological disciplines. One of the reasons for this might be the unclear nature of actual faking behavior. In the present study, we aimed to apply a modeling technique introduced by Ziegler, Maaß, Griffith, and Gammon (2015) that allows capturing of interindividual differences in faking behavior as a latent variable. We used this approach to isolate variance because of experimentally induced faking good and faking bad of the Big Five, and we predicted this variance with a variety of theoretically relevant constructs (socially desirable responding, overclaiming, and dark triad traits). We tested a sample (n = 233) divided between 2 experimental conditions and n = 167 persons in a control condition twice (honest/faking and honest/honest). The application of the modeling approach for all 5 personality domains was successful. In a second step, factor scores for all faking variables derived from these prior analyses were tested for homogeneity within each faking condition. Results showed that whereas faking was neither homogeneous within each condition (i.e., faking good vs. faking bad), nor was it homogeneous across conditions. Thus, faking is a complex psychological process that is responsive to specific situational demands. In a final step, the faking variables representing faking good and faking bad were regressed onto scores from other measures. The results indicated that the common variance shared by some social desirability scales predicted faking. We discuss theoretical and practical implications of these findings. (PsycINFO Database Record (c) 2019 APA, all rights reserved).