Most national surveys of health care utilisation capture only self-reported measures of morbidity. If self-reported morbidity is measured with error, then the results of applied work may be misleading. In this paper we propose a model of the relationship between morbidity and health service utilisation which allows for reporting errors and simultaneity. Errors in self-reported morbidity are expressed as a function of person-specific reporting thresholds and recent contact with health services, arising because of better self-evaluation of current health status or a desire to justify consumption of a publicly-provided good. We demonstrate the bias in ignoring the potential problems of reporting errors and simultaneity for a variety of special cases, but in the general case the biases are of ambiguous sign. The empirical nature of these biases is investigated using limiting long-standing illness (LLI) and recent contact with a General Practitioner (GP) in two waves of The UK Health and Lifestyle Survey. Biomedical measures of functioning are used as objective indicators of health status. We find evidence of substantial and significant differences between individuals in reporting thresholds and some evidence that the reporting of LLI may depend on recent visits to a GP. Adjustments for these biases significantly increase the estimated effect of morbidity on utilisation.