We are developing a CdZnTe pixel detector with a custom low- noise analog VLSI readout for use in the High-Energy Focusing Telescope balloon experiment, as well as for future space astronomy applications. The goal of the program is to achieve good energy resolution (< 1 keV FWHM at 60 keV) and low threshold in a sensor with approximately 500 micrometers pixels. We have fabricated several prototype detector assemblies with 2 mm thick, 680 by 650 micrometers pitch CdZnTe pixel sensors indium bump bonded a VLSI readout chip developed at Caltech. Each readout circuit in the 8 X 8 prototype is matched to the detector pixel size, and contains a preamplifier, shaping amplifiers, and a peak stretcher/discriminator. In the first 8 X 8 prototype, we have demonstrated the low-noise preamplifier by routing the output signals off-chip for shaping and pulse-height analysis. Pulse height spectra obtained using a 241Am source, collimated to illuminate a single pixel, show excellent energy resolution of 1.1 keV FWHM for the 60 keV line at room temperature. Line profiles are approximately Gaussian and dominated by electronic noise, however a small low energy tail is evident for the 60 keV line. We obtained slightly improved resolution of 0.9 keV FWHM at 60 keV by cooling the detector to 5 degree(s)C, near the expected balloon- flight operating temperature. Pulse height spectra obtained with the collimated source positioned between pixels show the effect of signal sharing for events occurring near the boundary. We are able to model the observed spectra using a Monte-Carlo simulation that includes the effects of photon interaction, charge transport and diffusion, pixel and collimator geometry, and electronic noise. By using the model to simulate the detector response to uncollimated radiation (including the effect of finite trigger threshold for reconstruction of the total energy of multi-pixel events), we find the energy resolution to be degraded by only 10% for full-face illumination, compared to the collimated case. The small value of the degradation is due directly to the low readout noise and amplifier threshold.