The precise ordering of the hamster retinocollicular projection is established over the first two postnatal weeks, coincident with developmental cell death. We have used quantitative retrograde labelling to define topographic precision in the early postnatal projection, to describe its refinement and to assess the contribution played by selective retinal ganglion cell death. The hamster's short gestation period allows the investigation of events occurring prenatally in other rodents. Discrete injections of fluorescent beads in the superior colliculus followed by isodensity contour analysis of labelled retinal cells reveals a dramatic decrease in the extent of retina labelled between postnatal days 2, 6 and 12 (P2, P6, P12): the 20% contour encloses 38.3%, 8.3% and 1.8% of the retina at these ages. Paired injections of two different tracers at variable rostrocaudal (R–C) separations at P2 produced complete overlap of label even when injections were separated by over 1 mm. This was not true for paired mediolateral injections at P2 that were separated by more than 500 µm. Analysis of the segregation of the two tracers (‘nearest-neighbour analysis’) shows topography improving with age so that by P12 injections separated rostrocaudally by more than 500 µm produced no overlap in the retina. To examine the contribution of selective ganglion cell death to topographic refinement, animals given paired R–C injections at P2 were allowed to survive until P12. Nearest-neighbour analysis reveals significantly more order in the P2–P12 retinae than after overnight survival. Thus, selective cell death plays a small but appreciable role in correction of topographical errors.