The interobserver and intraobserver agreement between two microscopists was independently assessed in a blind fashion to evaluate the reproducibility of the results of the quick diagnosis of Chlamydia trachomatis infection with direct immunofluorescence. The study was carried out in two phases, and 604 samples of cervical and urethral exudate from patients seen at a center for the prevention and therapy of sexually transmitted diseases were evaluated. The two observers had a kappa index of agreement between them of 0.41, with 95% confidence intervals (95% Cl) of 0.31-0.51, while their agreement within themselves was 0.78 (95% Cl: 0.65-0.91) for the first observer and 0.61 (95% Cl: 0.47-0.75) for the second one. After discussion of the data and analysis of the reasons of disagreement, the evaluation was repeated. In this second phase, both interobserver and intraobserver agreement were substantially improved. The kappa index was 0.71 (95% Cl: 0.61-0.80) for the former, whereas regarding the latter the first observer had a kappa index of 1 (95% Cl: 0.86-1.13) and the second observer an index of 0.82 (95% Cl: 0.61-0.80). It was concluded that the test may have problems in its practical performance if the microscopists are not previously trained with systematic programs for the evaluation of results.