A battery of cardiovascular reflex tests is usually performed for the diagnosis of autonomic neuropathy. The tests discriminate well between normal and definitely abnormal autonomic function. However, in some patients the results are borderline and their autonomic status cannot be better defined. We performed multivariate statistical analysis of six cardiovascular autonomic tests with the aim of increasing their diagnostic efficiency. Eighty-five healthy subjects and 95 patients at risk for autonomic neuropathy were studied. Principal component analysis and two pattern recognition methods, the Bayesian technique and the SIMCA method, were applied. It was found that: (1) normal models obtained by Bayesian analysis showed very high specificity and sensitivity; (2) a battery of two tests for parasympathetic function (R-R interval variation test, deep breathing) and two tests for sympathetic function (blood pressure responses to standing and to sustained handgrip) provide an appropriate diagnostic approach, if multivariate analysis is used; (3) multivariate analysis allows a more precisely defined assessment of autonomic nervous system function in so-called borderline patients.