The effect of cimetidine and placebo was examined in 123 patients with non-ulcer dyspepsia (NUD) by means of a 12-day multi-crossover model with 5 regular interchanges between cimetidine and placebo. The evaluation of effect in individual patients was based on the number of times cimetidine was associated with less symptoms than the preceding or following placebo period. If cimetidine had no effect, the probability of being defined as a cimetidine responder was 25%. In general, cimetidine was associated with less symptoms than placebo (p less than 0.0001). Forty patients were identified as cimetidine responders (R) and the remaining patients were termed non-responders (NR). Symptoms compatible with gastroesophageal reflux were significantly more frequent in R than in NR, whereas the opposite was true for symptoms of the irritable colon syndrome. The ability of symptoms selected by stepwise logistic regression to predict response to cimetidine showed at best a sensitivity of 75% and a specificity of about 65%. No differences were found between R and NR with regard to acid secretion, endoscopic and histologic findings, or the result of an acid perfusion test. The present study supports the existence of a subgroup of cimetidine responders among patients with NUD characterized by symptoms suggestive of gastroesophageal reflux disease in the absence of confirmatory objective evidence.