A novel neural network architecture is developed to study some basic aspects in early visual information processing in vertebrate retina, which is based on the neural anatomy and function of retinal neurons in tiger-salamanders and mudpuppies. The architecture comprises neural models of photoreceptors, horizontal cells, bipolar cells, amacrine cells, and ganglion cells. The main response characteristics of the retinal neurons are studied, and the model predictions are compared with the corresponding data. The possible action of $\gamma$-aminobutyric acid (GABA) inhibition from horizontal cells are also examined. The simulation results show that isolated or combined action of GABA_A and GABA_B can generate the dynamics observed in bipolar cells. The simulations removing the asymmetry in the inhibitory pathway produces results consistent with the hypothesis that the directional selectivity results from an asymmetrical inhibition. Model responses for directionally selective ganglion cells to moving spots are found to be in qualitative agreements with data from the turtle when tested with and without amacrine inhibition.