OBJECTIVE: To assess the intra-observer and overall agreement in the interpretation of chest X-rays (CXRs) performed for detecting tuberculosis (TB) among immigrants in Switzerland. METHOD: Four hundred digitalised CXRs from the files of immigrant registration centres were selected and read twice in random order by three readers. The readers had to assess (1) if the picture was normal or abnormal; (2) if an abnormality was suggestive of TB; and (3) if the suspicion of TB needed an immediate examination (potentially smear-positive TB). The intra-observer and overall agreements were expressed as kappa with standard error. RESULTS: Due to losses for technical reasons, 377 of the 400 pictures were analysed. The intra-observer agreement was 0.39-0.90 for any abnormality, and 0.60-0.82 for TB needing an urgent examination. The overall agreements were: 0.55 (all three readers) and 0.84 (two best readers) for any abnormality, and 0.64 (all three readers) and 0.80 (two best readers) for active TB. CONCLUSIONS: The intra-observer and overall agreements for the detection of abnormalities on digitalised CXRs and for the presence of possible active TB depend on the reader's experience. It was good between experienced readers and fair between and within the inexperienced reader.