A multicentric, prospective phase III study was carried out with the aim of testing the so-called 'worst drug rule' hypothesis, which suggests the use of an effective but 'less active' regimen that first eradicates tumoral cells resistant to a second effective and 'more active' regimen. With respect to this hypothesis, we considered the cisplatin plus vinorelbine regimen (CCDP/VNR) as the more active regimen compared with the non-cisplatin-containing regimen of ifosfamide plus high-dose epirubicin (IFO/EPI). Thus, a randomized study was carried out to compare the sequencial strategy of three cycles of CDDP/VNR followed by three cycles of IFO/EPI with the opposite sequence in advanced non-small-cell lung cancer. A total of 100 consecutive previously untreated patients with stage III-IV non-small-cell lung cancer were centrally randomized in two arms according to stage of disease and the performance status. Patients allocated to arm A received CDDP (100 mg m-2 on day 1) plus VNR (25 mg m-2 i.v. on days 1 and 8) every 21 days for three cycles (step 1) followed, after restaging, by three cycles of IFO (2.5 g m-2 with mesna on day 1) plus high-dose EPI (100 mg m-2 on day 1) every 21 days (step 2). Patients in arm B received the opposite sequence. Type and rates of objective response were evaluated after step 1 and step 2 in agreement with WHO criteria and an intent-to-treat analysis. Patients were also analysed for toxicity patterns, time to progression and survival. After the first three cycles (step 1), overall response rate (ORR), calculated according to an intent-to-treat analysis, was 47% and 21% for arm A and arm B respectively (P = 0.0112). ORR for stage III patients was 55% and 14% for arm A and B respectively (P = 0.0097). In stage IV patients ORR was higher in arm A than in arm B (42% vs 28%) but not statistically significant (P = 0.4). Clinical responses to the shift of chemotherapy (step 2) showed that no patient pretreated with CDDP/VNR and subsequently treated with IFO/EPI showed further response, whereas in the inverse sequence arm CDDP/VNR was able to induce 26% partial response (PR) rate in patients pretreated with IFO/EPI. This difference was statistically significant (P = 0.037). The overall median time to progression (TTP) of arm A and arm B did not significantly differ (6 vs 4 months; P = 0.665). However, median TTP of stage III patients was, respectively, 7 months for arm A and only 3 months for arm B. This difference was statistically significant (P = 0.049). Median overall survival (OS) was 9 and 7 months respectively for arm A and arm B. Despite this trend the difference was not significant (P = 0.328). Median OS of stage III patients showed a statistically significant advantage for arm A over arm B (13 vs 7 months, P = 0.03). In addition, no statistically significant difference in OS was recorded for stage IV patients (both arms 7 months, P = 0.526). Our data do not confirm Day's 'worst drug rule' hypothesis, at least in patients with advanced non-small-cell lung cancer treated with the above-mentioned regimens. The combination of CDDP and VNR seems more active, at least in terms of response rate, than the IFO/EPI, which performed poorly.