Rats received bilateral lesions of the nucleus basalis magnocellularis by infusion of biotenic acid. Two weeks after the lesion, a suspension of genetically modified primary rat fibroblasts was grafted dorsal to the nucleus basalis magnocellularis (2 x 10(5) cells per side). The fibroblasts were either infected with the gene for human beta-nerve growth factor or Escherichia coli beta-galactosidase. The nerve growth factor-producing fibroblasts released 67 ng nerve growth factor/10(5) cells per day in vitro. Two weeks after implantation of the fibroblasts, spatial learning was tested in the Morris water-maze. Nerve growth factor-producing fibroblasts, but not beta-galactosidase-producing fibroblasts ameliorated the deficit in acquisition of the water-maze task. In addition, spatial acuity was improved to near-normal levels by the nerve growth factor-producing grafts. Choline acetyltransferase activity in cortical areas and hippocampus was not affected by the nerve growth factor-producing grafts. Both grafted groups showed a similar reduction in the level of dopamine, but not homovanillic acid or 3-methoxytyramine, in the frontal cortex. Levels of norepinephrine, epinephrine and serotonin and their metabolites in the neocortex and hippocampus were not affected by the lesion or the grafts. Nerve growth factor-producing grafts increased the size of remaining nerve growth factor-receptor (p75) immunoreactive neurons in the nucleus basalis magnocellularis by 25%. Nucleus basalis magnocellularis lesions reduced the integrated optic density of choline acetyltransferase-positive fiber staining in the ventral neocortex by 46%, but nerve growth factor-producing grafts restored this area to 86% of control. These data suggest that nerve growth factor-producing grafts can cause a marked behavioral improvement, probably through the partial restoration of the lesioned projection from nucleus basalis magnocellularis to neocortex.