Background Although depression and mania are often assumed to be polar opposites, studies have shown that, in patients with bipolar disorder, they are weakly positively correlated and vary somewhat independently over time. Thus, when investigating relationships between specific psychological processes and specific symptoms (mania and depression), co-morbidity between the symptoms and changes over time must be taken into account. Method A total of 253 bipolar disorder patients were assessed every 24 weeks for 18 months using the Hamilton Rating Scale for Depression (HAMD), the Bech–Rafaelsen Mania Assessment Scale (MAS), the Rosenberg Self-Esteem Questionnaire (RSEQ), the Dysfunctional Attitudes Scale (DAS), the Internal, Personal and Situational Attributions Questionnaire (IPSAQ) and the Personal Qualities Questionnaire (PQQ). We calculated multilevel models using the xtreg module of Stata 9.1, with psychological and clinical measures nested within each participant. Results Mania and depression were weakly, yet significantly, associated; each was related to distinct psychological processes. Cross-sectionally, self-esteem showed the most robust associations with depression and mania: depression was associated with low positive and high negative self-esteem, and mania with high positive self-esteem. Depression was significantly associated with most of the other self-referential measures, whereas mania was weakly associated only with the externalizing bias of the IPSAQ and the achievement scale of the DAS. Prospectively, low self-esteem predicted future depression. Conclusions The associations between different self-referential thinking processes and different phases of bipolar disorder, and the presence of the negative self-concept in both depression and mania, have implications for therapeutic management, and also for future directions of research.