The primary aim of our study was to demonstrate how the diagnostic characteristics of skin biopsy used to evaluate small fiber involvement in patients with different causes of polyneuropathy are intrinsically related to the method used to establish the reference values (cut-off values). We also investigated intraepidermal nerve fiber (IENF) density and abnormalities in quantitative sensory testing (QST) in patients with different causes of polyneuropathy and signs of small fiber involvement. A total of 210 patients with symptoms and signs of polyneuropathy were entered into the study. All patients underwent neurological examination, nerve conduction studies, QST on the thigh and distal part of the calf with detection of warm and cold perception thresholds, and skin biopsy with assessment of IENF density. Cut-off values for IENF density were established from our reference material using Z-scores (calculated from multiple regression analysis), fifth percentile, and receiver operating characteristic (ROC) analysis. Of the patients participating in the study, 65 had an established diagnosis of diabetes mellitus, 70 were classified with idiopathic polyneuropathy, and 75 had other possible causes of polyneuropathy. Forty-five patients met the criteria for small fiber polyneuropathy (SFN), and the remaining 165 had also involvement of large nerve fibers. Of the total patient cohort, 84 (40%) had reduced IENF density based on the Z-score, and 106 patients (50%) had at least one abnormality based on QST. In the SFN group, skin biopsy showed a sensitivity of 31% and a specificity of 98% when reference values were presented with Z-scores. When the fifth percentile was used as the cut-off value (6.7 fibers/mm), sensitivity was 35% and specificity 95%. Applying the ROC analysis with a chosen sensitivity of 78% and specificity of 64%, we had a cut-off point of 10.3 fibers/mm. We conclude that skin biopsy with assessment of IENF is a useful method for investigating patients with SFN. The diagnostic value of the test, however, depends upon on the approach used to estimate the reference values.