Knowledge about typical change patterns of psychotherapy patients can help to improve treatment guidelines for psychological disorders. Recent studies showed that it is possible to identify several patient subgroups with regard to their early change pattern. However, although focusing on the early phase of treatment, change patterns in later stages have hardly been investigated yet. In this study, Growth Mixture Modelling was used to identify latent change classes in different phases of therapy in a naturalistic sample of 1229 psychotherapy outpatients. Furthermore, this paper inquired into the relation between the change patterns in different phases as well as their predictive power for therapy length and outcome. Results revealed different change patterns for the three investigated phases. While in an early treatment phase, (sessions one–six) five different change patterns could be identified: the number of change classes decreased considerably over time, resulting in three patterns in the second (sessions 7–12) and two in the third phase (sessions 13–18). In each phase, by far, the biggest class showed a pattern of good progress with small/no further improvements. Copyright © 2013 John Wiley & Sons, Ltd. Key Practitioner Messages Most change in patients' progress estimates took place in an early phase of the treatment and levelled out on a relatively high level in later phases of the treatment. Substantial improvements were still present in later phases of the treatment but occurred less frequent than in early stages. Continuous outcome monitoring and feedback systems should integrate progress measures to monitor patients progress especially in the early phase of the treatment and feed the so gained information back to therapists.