The use of observer-rated scales requires that raters be trained until they have become reliable in using the scales. However, few studies properly report how training in using a given rating scale is conducted or indeed how it should be conducted. This study examined progress in interrater reliability over 6 months of training with two observer-rated scales, the Cognitive Errors Rating Scale and the Coping Action Patterns Rating Scale. The evolution of the intraclass correlation coefficients was modeled using hierarchical linear modeling. Results showed an overall training effect as well as effects of the basic training phase and of the rater calibration phase, the latter being smaller than the former. The results are discussed in terms of implications for rater training in psychotherapy research.