The diagnosis of mycosis fungoides (MF) is notoriously difficult to establish because in the early stages, histological features may be nonspecific or merely suggestive. To standardize the diagnosis of MF. We studied 138 patients with suspected MF referred over a 7-year period to a university department of a dermatology-based cutaneous lymphoma clinic. Six diagnostic criteria were evaluated: clinical morphology, clinical distribution, skin biopsy T-cell receptor gene rearrangement (TCR-GR), skin biopsy pan T-cell marker loss > or = 2, skin biopsy CD4/CD8 ratio > or = 6, and skin biopsy diffuse epidermal HLA-DR expression. These six clinical and laboratory criteria were compared by logistic regression analysis in patients with histologically diagnosed MF and those with benign disease. Of the 138 patients, 74 had histology of MF, 47 of benign dermatoses and 17 were indeterminate. Close associations were found between a histological diagnosis of MF and TCR-GR (odds ratio 14.4), classical morphology (7.5), classical distribution (2.5) and diffuse epidermal HLA-DR expression (2.8). Logistic regression models were developed depending on the availability of data (either TCR-GR or HLA-DR). Probabilities for correctly diagnosing MF compared with histology as the 'gold standard' were derived from these logistic regression models. A scoring system assigning point values based on these probabilities was then created in order to assist the clinician in making the diagnosis. If using TCR-GR data, a positive TCR-GR = 2.5 points, the presence of classical morphology = 2.0 points, and the presence of classical distribution = 1.5 points. A total score of > or = 3.5 points assigns a high probability (> 85%) of having MF. If using HLA-DR expression, then the presence of classical morphology = 2.5 points, a positive diffuse epidermal HLA-DR expression = 2.0 points, and the presence of classical distribution = 1.5 points. In this case, a total score of > or = 4.0 points assigns a high probability (> 85%) of MF. The logistic regression models and scoring systems integrate clinical and laboratory assessments, allow rapid probability estimation, and provide a threshold for the diagnosis of MF in an objective, standardized manner.