From recent studies of locomoting fish keratocytes it was proposed that the dynamic turnover of actin filaments takes place by a nucleation-release mechanism, which predicts the existence of short (less than 0.5 microns) filaments throughout the lamellipodium (Theriot, J. A., and T. J. Mitchison. 1991. Nature (Lond.). 352:126-131). We have tested this model by investigating the structure of whole mount keratocyte cytoskeletons in the electron microscope and phalloidin-labeled cells, after various fixations, in the light microscope. Micrographs of negatively stained keratocyte cytoskeletons produced by Triton extraction showed that the actin filaments of the lamellipodium are organized to a first approximation in a two-dimensional orthogonal network with the filaments subtending an angle of around 45 degrees to the cell front. Actin filament fringes grown onto the front edge of keratocyte cytoskeletons by the addition of exogenous actin showed a uniform polarity when decorated with myosin subfragment-1, consistent with the fast growing ends of the actin filaments abutting the anterior edge. A steady drop in filament density was observed from the mid-region of the lamellipodium to the perinuclear zone and in images of the more posterior regions of lower filament density many of the actin filaments could be seen to be at least several microns in length. Quantitative analysis of the intensity distribution of fluorescent phalloidin staining across the lamellipodium revealed that the gradient of filament density as well as the absolute content of F-actin was dependent on the fixation method. In cells first fixed and then extracted with Triton, a steep gradient of phalloidin staining was observed from the front to the rear of the lamellipodium. With the protocol required to obtain the electron microscope images, namely Triton extraction followed by fixation, phalloidin staining was, significantly and preferentially reduced in the anterior part of the lamellipodium. This resulted in a lower gradient of filament density, consistent with that seen in the electron microscope, and indicated a loss of around 45% of the filamentous actin during Triton extraction. We conclude, first that the filament organization and length distribution does not support a nucleation release model, but is more consistent with a treadmilling-type mechanism of locomotion featuring actin filaments of graded length. Second, we suggest that two layers of filaments make up the lamellipodium; a lower, stabilized layer associated with the ventral membrane and an upper layer associated with the dorsal membrane that is composed of filaments of a shorter range of lengths than the lower layer and which is mainly lost in Triton.