This study evaluated the reproducibility of nerve function assessment in a group of 132 diabetic patients with moderate peripheral polyneuropathy. Patients were investigated at the beginning and the end of the run-in period (a 1-month placebo period) of a multicentre trial of an aldose-reductase inhibitor (Ponalrestat). Reproducibility was evaluated by performing four types of tests: quantitative visual scales of symptoms, quantitative sensory assessment (vibration perception thresholds in medial malleolus and great toe, foot thermal perception threshold to hot and cold), electrophysiological investigations on the dominant side (conduction velocities and potential amplitudes of sensory and median motor nerve, sural and peroneal nerves, amplitudes of F waves of median motor and peroneal nerves) and cardiac autonomic tests (Valsalva, deep-breathing, lying-to-standing). Reproducibility was poor for symptoms, thermal sensitivity, and potential amplitudes. It was satisfactory (total coefficient of variation < 50%) for all the other parameters and even very good (total variation coefficient < 26%, intra-subject variation factors corresponding to < 56% of total variance) for velocities of sensory and median motor and peroneal nerves, the amplitudes of F waves and the three autonomic tests. For most of the parameters total variance was mainly related to inter-subject variability. However, inter-subject variability for the three cardiac autonomic tests was very low and at least one cardiac autonomic test was altered in all the patients. Inter-centre variability was low for all the parameters, except for action potential amplitudes and for F wave velocity of the median motor nerve. This study suggests those parameters that are appropriate for the assessment of diabetic neuropathy and for therapeutic trials.(ABSTRACT TRUNCATED AT 250 WORDS)