Early therapeutic alliance is usually measured by the rating of a single session (between the third and the fifth sessions). However, there is a strong argument in favor of viewing early alliance as a developing process. This study examined the relationship between patient's rating of the helping alliance (HAq) at each session and therapy outcome. This comparison was repeated using patterns of alliance over the course of treatment. Patterns of therapeutic alliance development were detected by clustering ratings of a sample of N = 70 outpatients across four sessions of very brief psychotherapeutic intervention. Cluster analysis revealed two main patterns (shapes) of alliance development: (i) stable alliance, and (ii) linear growth pattern. These patterns are more predictive of symptom improvement and social adjustment than single ratings, whereas single ratings measuring the strength of alliance are more correlated with patient's satisfaction.