The purpose was to test whether physicians can validly and reproducibly diagnose diabetic sensorimotor polyneuropathy (DSPN). Twelve physicians assessed 24 patients with diabetes mellitus (DM) on consecutive days (576 examinations) with physical features and voice disguised. Results were compared to gold standard 75% group diagnosis (dx) and a nerve conduction score (Sigma5 NC nds). Masking of patients was achieved. Reproducibility measured by the kappa coefficient and compared to Sigma5 NC nd varied considerably among physicians: median and ranges: signs 0.8 (0.32-1.0); symptoms 0.79 (0.36-1.0), and diagnoses 0.47 (0.33-0.84), both low and high scores indicating poor performance. There was substantial agreement between 75% group dx and confirmed NC abnormality (abn). As compared to Sigma5 NC, individual physicians' clinical dx was excessively variable and frequently inaccurate. Study physician dx from signs and symptoms were excessively variable, often overestimating DSPN. Specific approaches to improving clinical proficiency should be tested.