Blastomere lineages are differentially biased to produce different neurotransmitter subtypes of amacrine cells (Huang and Moody, 1995, 1997,). To elucidate when this bias is acquired, we examined amacrine lineages at different early developmental times. Our experiments demonstrate that the bias to express dopamine and neuropeptide Y amacrine fates involves several steps before the formation of the definitive optic cup. At cleavage stages, a retinal progenitor that contributes large numbers of cells is already biased to produce its normal repertoire of dopamine amacrine cells, as revealed by transplantation to a new location, whereas the amacrine fate of a progenitor that contributes fewer cells is modified by its new position. At neural plate stages, not all retinal progenitors are multipotent. Nearly one-half populate only the inner nuclear layer and are enriched in amacrine cells. During early optic vesicle stages, an appropriate mitotic tree is required for dopamine and neuropeptide Y, but not serotonin, amacrine cell clusters to form. Thus, the acquisition of amacrine fate bias involves intrinsic maternal factors at cleavage, fate restriction in the neural plate, and specified mitotic patterns in the optic vesicle. At each of these steps only a subset of the embryonic retinal progenitors contributing to amacrine subtypes is biased; the remaining progenitors maintain multipotency. Thus, from the earliest embryonic stages, progenitors of the retina are a dynamic mosaic. This is the first experimental demonstration of amacrine fate decisions that occur during early embryonic periods in advance of the events described in the later, committed retina.