Background: Bipolar disorder is a common chronic illness characterized by high levels of morbidity and all-cause mortality. Lithium is one of the gold standard mood stabilizer treatments, but the identification of good, partial and non-responders in clinical settings is inconsistent. Methods: We used an established rating scale (the Alda scale) to classify the degree of lithium response (good response, partial response, non-response) in a large, multicentre clinically representative sample of well-characterized cases of bipolar disorders I and II. Next, we examined previously reported clinical predictors of response to determine which factors significantly differentiated between the three response groups. Results: Of 754 cases, 300 received lithium, for at least 6 months, as a treatment for bipolar disorder (40%). Of these cases, 17% were classified as good response, 52% as partial response and 31% as non-response. Lifetime history of mixed episodes ( p = 0.017) and alcohol use disorders ( p = 0.015) both occurred in >20% of partial response and non-response groups but <10% of good response cases. Family history of bipolar disorder I was of borderline statistical significance, being more frequent in the good response group (38%) compared with the non-response group (18%). There was a trend ( p = 0.06) for bipolar disorder II to be associated with non-response. Conclusions: Only three factors previously identified as predictors of lithium response significantly differentiated the response groups identified in our sample. Interestingly, these factors have all been found to co-occur more often than expected by chance, and it can be hypothesized that they may represent a shared underlying factor or dimension. Further prospective studies of predictors and the performance of the Alda scale are recommended.