Serial peak expiratory flow (PEF) records are useful for the screening and diagnosis of patients with occupational asthma. We have presented work on a method based on linear discriminant analysis (OASYS-2) and have now tried to improve on this technique and investigate the repeatabilities of expert interpretation. 268 serial PEF records made by workers with possible occupational asthma were divided into four sets. The first two were development sets; development set 1 was used to develop a discriminant analysis to match the human expert, who scored each complex (a sequence of work days, then rest days, then work days, or its counterpart: rest-work-rest) for its likelihood of having a work-related effect. This was modelled using 729 new measurements from each complex, together with the 50 measurements originally used in the development of OASYS-2. Linear discriminant analysis is a statistical technique which can refine from a large number of indices a narrower range that best predict a known outcome. Combinations of these indices are weighted and used to assign outcomes for novel cases. Models were produced containing a combination of measurements that generated scores best matching the expert. Development set 2 was used to test the model. Sets 3 and 4 were "gold standard" sets where the diagnosis had been made independently of the PEF record. Set 3 was used to set the cut-off for an occupational effect, the sensitivity and specificity for the combined model was determined from the fourth (gold standard) set. The fourth set was also used to determine the sensitivity and specificity of the human expert. The repeatability of the human expert re-scoring the same complexes had a weighted kappa score of 0.71. OASYS-3 was 100% repeatable. Comparing the scores awarded to whole records by the new (OASYS-3) and OASYS-2 analysis methods with the scores awarded by the human expert revealed mean (95% CI) differences of -0.28 (0.30, -0.26) and -0.34 (-0.37, -0.31) respectively. Hence both OASYS-3 and OASYS-2 tended to score records less positively for work-related changes in PEF than the expert. OASYS-3 scored complexes marginally better than OASYS-2. The sensitivity of OASYS-3 was better than OASYA-2 (82% and 75% respectively) for an equivalent specificity (94%). The sensitivity of the human expert was 100% with a specificity of 93%. OASYS-3 provides an objective method of interpreting serial peak flow records with a sensitivity and specificity approaching that of a human expert and is a modest improvement on OASYS-2.