Background Coordinated Anxiety Learning and Management (CALM) is a model for delivering evidence‐based treatment for anxiety disorders in primary care. Compared to usual care, CALM produced greater improvement in anxiety symptoms. However, mean estimates can obscure heterogeneity in treatment response. This study aimed to identify (1) clusters of participants with similar patterns of change in anxiety severity and impairment (trajectory groups); and (2) characteristics that predict trajectory group membership. Methods The CALM randomized controlled effectiveness trial was conducted in 17 primary care clinics in four US cities in 2006–2009. 1,004 English‐ or Spanish‐speaking patients age 18–75 with panic, generalized anxiety, social anxiety, and/or posttraumatic stress disorder participated. The Overall Anxiety Severity and Impairment Scale was administered repeatedly to 482 participants randomized to CALM treatment. Group‐based trajectory modeling was applied to identify trajectory groups and multinomial logit to predict trajectory group membership. Results Two predicted trajectories, representing about two‐thirds of participants, were below the cut‐off for clinically significant anxiety a couple of months after treatment initiation. The predicted trajectory for the majority of remaining participants was below the cut‐off by 9 months. A small group of participants did not show consistent improvement. Being sicker at baseline, not working, and reporting less social support were associated with less favorable trajectories. Conclusions There is heterogeneity in patient response to anxiety treatment. Adverse circumstances appear to hamper treatment response. To what extent anxiety symptoms improve insufficiently because adverse patient circumstances contribute to suboptimal treatment delivery, suboptimal treatment adherence, or suboptimal treatment response requires further investigation.