There are several predictors of severe hypoglycaemia in patients with insulin-dependent diabetes mellitus (IDDM), many of which are correlated. To assess factors which may be predictive of severe hypoglycaemia, structural equation modelling was used to test specific hypotheses using prospective data. Sixty patients with insulin-dependent diabetes mellitus (IDDM) were studied prospectively for one year during which any episodes of severe hypoglycaemia, asymptomatic biochemical hypoglycaemia, and glycaemic control were documented. Half the patients reported hypoglycaemia unawareness and they were matched for HbA1 with the rest. Baseline measurements included symptomatic awareness of hypoglycaemia, fear of hypoglycaemia, previous history of hypoglycaemia, glycaemic control, and peripheral autonomic function. Formal structural equation modelling was performed on these variables and a model was constructed that expressed the putative causal associations among the variables. The frequency of severe hypoglycaemia (measured prospectively) correlated significantly with duration of diabetes, awareness of hypoglycaemia, patient's age, history of previous severe hypoglycaemia and autonomic function scores. HbA1 did not show significant correlation, possibly because of the narrow range in the subject population. In the structural equation modelling exercise, at least 18% of the variance of severe hypoglycaemia, measured prospectively, was accounted for by a history of severe hypoglycaemia, the state of awareness of hypoglycaemia, and the autonomic function score. Over 25% of the variance of 'worry' on the hypoglycaemia fear scale was accounted for by a history of previous severe hypoglycaemia. An assessment of multiple risk factors for hypoglycaemia may be of value in advising individual patients about their diabetes care.