The main purpose of this study was to analyse the structure and content of real-life causal attributions, using this data to examine some important theoretical and empirical issues within attribution theory. Chapter one provides a justificatory backdrop to this research. It is argued here that this type of real-life exploratory research is sorely needed because the vast bulk of the attribution research has hitherto utilised a laboratory based experimental approach which has weak ecological validity. The causal attributions examined came from free-response verbal protocols of explanations for marital separation. The sample comprised 29 males and 33 females, living in Hamilton, New Zealand, all of whom had been physically separated for less than 18 months. Two attribution taxonomies were developed to classify the attributions, both of which had adequate inter-rater reliability. One taxonomy was in terms of content attributions (e.g. communication, sex, etc.), while the other was an abstract taxonomy in terms of attitudes, personality characteristics, etc. The findings from these taxonomies, which included some interesting sex differences, are fully discussed. The technique of subjective magnitude estimation (developed by S.S. Stevens within the field of psychophysics) was used by respondents to give causal importance weights to the attributions derived from the verbal protocols. Ratio-level data, representing perceived causal responsibility for different attribution targets (self, ex-spouse, external factors, etc.) were derived from these weights, and multiple regression tests were carried out to examine the impact exerted by a range of independent variables on these attribution category percentages. It was found that the most influence on these derived attribution percentages was exerted by which partner decided to leave, education level and personal control. There was also some evidence that attribution patterns were related to self-reported levels of coping. A range of evidence, including convergent validity and reliability correlations, is presented which supports the validity of using this technique for deriving ratio level data from verbal protocols. The descriptive and other data gathered was found to be both consistent and inconsistent with different aspects of attribution theory. The major findings inconsistent with the extant theories, were the range and complexity of people’s explanations, and people’s tendency to make attributions for dispositions. Conversely, the data strongly supported the psychological validity of the internal-external distinction, and also supported attribution theory’s assumption that personal dispositions are the most important class of attribution. A number of other empirical attribution issues are examined, the main ones being the self-serving defensive attribution model vs the information processing model and the question of where the causal attributions come from. In the former case it is argued that both models are supported by our data. In the latter case, it is concluded that, (a) there does exist a socially shared theory of what causes marriages to break up, (b) that separated people do not select their attributions by consulting this theory, and (c) that they do cognitively access this theory in the process of providing causal importance weights for the attributions. A range of other findings is presented. For example, one finding of interest was that more females (64%) decided to leave the marriage than males (21%). In all cases explanations are proffered and discussed for the reported findings.