The repeatability of the bronchoalveolar lavage (BAL) was assessed prospectively in 44 mechanically ventilated patients with suspected nosocomial pneumonia. Two BAL were performed in the same lung area (contiguous segment) during two fibroscopic procedures performed with a thirty minute interval. All the bronchoscopies were performed by the same operator. The statistical analysis looked out for bias (MacNemar test), agreement, and repeatability (kappa test). In the 44 patients studied, the qualitative repeatability (i.e., presence or absence of bacteria) was excellent (95.4%). However, in the 16 patients having at least one positive culture, these results were more controversial. The quantitative repeatability for bacteria (same log10 for both BAL of the same patient) was the lowest of all the results (26.7%). The distinction between presence and absence of bacterial pneumonia (based on the 10[4] cfu/ml threshold) showed a repeatability of 75% with no bias, an agreement of 47% and a just-significant kappa test (test = 1.97; p = 1.96 for a 5% risk error). BAL seems to have excellent repeatability when sterile. Its repeatability when positive needs further studies to be assessed.