Previous studies of oral physostigmine in the treatment of Alzheimer's disease have: (1) assumed physostigmine is effective only in mildly affected patients; (2) relied on an initial "dose-finding" phase to determine the most effective dose and excluded nonresponders; and (3) primarily assessed memory. We examined the response of 22 patients to six different daily dosages of oral physostigmine, using selective reminding tests that were administered twice daily. Nine patients had a "best" dose/day (mode = 13 mg/day), which was used in a subsequent double-blind crossover study. The other 13 were given the highest tolerated dose. The selective reminding test and a full neuropsychological battery were given during the drug and placebo periods. As a group, the 22 patients improved significantly on the Wechsler Adult Intelligence Scale-Revised Digit Symbol subtest and a shape cancellation task (p less than 0.05). Nine patients showed improved performance on the selective reminding test during physostigmine treatment, and 9 showed no response; 4 patients performed better during placebo treatment. Dose finding did not help in predicting response in the crossover study; only 2 of the 9 who showed improvement had a best dose. Dementia severity did not predict crossover response. This suggests that: (1) physostigmine as administered had no pronounced effect on memory in Alzheimer's disease; (2) oral physostigmine produces no greater benefits on memory in mildly than in moderately demented patients; (3) response in a dose-finding phase does not predict response in double-blind crossover; and (4) Digit Symbol and cancellation tasks may be more sensitive than memory tests to the effects of oral physostigmine.