Two groups of five pigeons each were trained to discriminate morphine 5.6 mg/kg vs cocaine 5.6 mg/kg (High dose cocaine group), and morphine 5.6 mg/kg vs cocaine 3.0 mg/kg (Low dose cocaine group), respectively. Within both groups, cocaine dose generalization gradients were radically flatter in comparison to those obtained when the same cocaine dosages were discriminated from saline, whereas morphine gradients, analogously compared, were only moderately so. Responding to non-drug tests did not deviate significantly from random in either group. After lowering morphine training doses to 3.0 mg/kg in a systematic replication of the experiment, training drugs generalization gradients took symmetrical, flattened, shapes in the High dose cocaine group, whereas in the Low dose cocaine group the previous relation between gradients was enhanced. Responding to non-drug tests in this phase did not deviate significantly from random in the High dose cocaine group, whereas the reverse was true with the Low dose cocaine group. Tests with novel drugs (apomorphine, LSD, pentobarbital and delta 9-THC), did not differentiate between the groups. The roles of generalization gradients, non-drug tests and novel drugs tests as measures of relative stimulus control of training stimuli in drug discrimination experiments are discussed within the framework of a drug discrimination model.