Research into test-retest stability of emotional and cognitive responses has focused on behavioural and ERP components in relation to very well-established paradigms (such as oddball, mismatch negativity, continuous performance task, etc.). However, there is a lack of evidence for more integrative paradigms that contain a variety of tasks in a single paradigm. The aim of the present study was to investigate the test-retest reliability for behavioural and ERP components (cue-locked P300 and task-locked P600) in a paradigm composed of three different error-recognition tasks: lexical, arithmetic, and algebraic. Participants completed the paradigm on two occasions, two weeks apart on average. We examined the between-session correlations and mean-level differences. The results showed moderate-to-strong between-session stability for accuracies (r = [.29-.69]), reaction times (r = [.48-.81]), and ERP P300 and P600 responses (r = [.70-.79] and r = [.45-.85], respectively). Overall, results indicated that there was more homogeneity of the estimates at the ERP level than in stability estimates at the behavioural level. The study highlights the importance of studying neural mechanisms involved in emotional and cognitive processing in addition to studying behavioural output.