Techniques are provided for reducing program disturb in a memory device. The techniques include compensating for a temperature in the memory device to reduce the upshift in the threshold voltage (Vth) of erased-state memory cells. A minimum allowable program pulse duration increases with temperature to account for an increase in the attenuation of a program pulse along a word line. A program pulse duration which accounts for reduced channel boosting at relatively high temperatures is reduced as the temperature increases. An optimum program pulse duration is based on the larger of these durations. The optimum program pulse duration can also be based on factors such as a measure of program disturb or a memory hole width. Program disturb can also be reduced by easing the requirements of a verify test for the highest data state.