Clinical trials have shown that behavioural treatments based on variants of prolonged-speech (PS) are best practice for reducing the stuttering rate in adults. However, while stuttering is significantly reduced or eliminated for most adults in the short-term with such treatment, relapse in the longer-term is common. Consequently, there has been interest in attempting to establish variables that predict responsiveness to PS-based treatments. Identifying such variables would enable risk prediction and also contribute to the tailoring of treatments to suit individuals who are less likely to benefit in the long-term from the conventional PS-based treatments. Variables that have been investigated to date are the stuttering severity, the attitude to communication and the locus of control.The present study revisited this issue with methodological improvements.Prospective, continuous measures of outcome several years after treatment, and well-powered, least-squares multivariate regression, with backwards elimination, were used to determine the best model to predict short- and long-term outcomes of a PS-based treatment. Other predictor variables were also included. The stuttering rates of 78 participants were measured immediately after the intensive stage of treatment, and during a surprise telephone call 3.5-5 years after treatment.Attitude to communication and locus of control were separate constructs to stuttering rate. While the stuttering rate predicted outcome, neither of the two non-behavioural variables predicted outcome in either the short- or the long-term.The most significant finding is that the non-behavioural variables of attitude to communication and locus of control did not predict treatment outcomes. This is contrary to the findings of most previous studies.