Working alliance has shown a predictive value of the outcome in different therapeutic settings but was not yet studied in a non-medical setting. The predictive value and the factor structure of the Working Alliance Inventory (WAI) [36-item client version; as reported by Horvath and Greenberg (J Couns Psychol 36:223–233, 1989)] were studied in a French primary care setting on a sample of 130 adults accessing social services. The WAI total score completed after the first meeting was positively predictive of quality of the working alliance 4 months later. An exploratory factor analysis produced two orthogonal factors which explained 45.12% of the total variance: a first factor (23 items) labelled ‘positive expectations about the usefulness of help’ and a second factor (13 items) labelled ‘absence of suspicion about the effects of help’. The WAI is shown to be applicable to primary care social work settings to measure the working alliance phenomenon and predict disruption of practitioner–client relationship.