The long slender retinal cones of fishes shorten in the light and elongate in the dark. Light-induced cone shortening provides a useful model for stuying nonmuscle contraction because it is linear, slow, and repetitive. Cone cells contain both thin (actin) and thick (myosinlike) filaments oriented parallel to the axis of contraction. This study examines the polarities of the cone's thin filaments and the changes in filament distribution which accompany light-induced contraction, in an attempt to elucidate the structural basis for the cone's contractile process. The proximal half of the cone is fixed to its cellular neighbors in the outer nuclear layer while the distal half is free. Thus, all shortening takes place in a necklike region (the myoid) in the distal half of the cone which extends into the space between the neural retina and the pigmented retinal epithelium. Thin filaments are found throughout the length of the cone, whereas thick filaments occur predominantly in the proximal (axon) regions of both light- and dark-adapted cones. Thus, thick filaments are primarily localized outside the region where shortening takes place. Observations from myosin subfragment-1 binding studies suggest that the cone's thin filaments are organized into two opposing sets. In the distal half of the cone (including the myoid), virtually all filaments have proximally directed arrowheads. In the more proximal regions of the axon, many thin filaments have opposite polarity, their arrowheads being distally directed. Near the synaptic proximal end of the light-adapted (contracted) cone, filaments of opposite polarities occur in approximately equal numbers. Thus, in the cone axon there appear to be two overlapping sets of actin filaments whose opposite polarities correspond to the two actin halves of a muscle sarcomere. In elongated, dark-adapted cones, thick filaments are localized throughout the axon region of the cone. In light, thick filaments accumulate towards the proximal end of the cone. These observations are consistent with a "sliding hypothesis" for cone contraction, in which thick myosinlike filaments produce sliding interdigitation of the two sets of oppositely directed actin filaments in the proximal axon region. Thus, the myoid thin filaments would be essentially reeled into the axon region to produce shortening. The mechanism of re-elongation depends on microtubules, as discussed in the companion paper.