Abstract In a pilot utility direct load control (DLC) program in New Jersey, air conditioners of volunteer household were remotely cycled in order to reduce electric load demand. The resulting load savings and comfort effects of the program are analyzed. A discomfort-from-cycling index is constructed from participant logs of cooling adequacy. Load reductions are calculated from 5-min consumption data. Based on both the discomport index and reported internal temperatures, the DLC cycling did not create a comfort problem for most participants. When the DLC equipment was activated, participants reported only slight increases in temperature (from 78.4 to 79.0°F) and discomfort (increasing from 7% to 15%). Also, participants achieving higher load savings were not significantly more likely to report discomfort. Two variables which did significantly predict load savings were duty cycle and frequency of thermostatic cycling. Current DLC equipment uses a preset cycle-off time, typically 25% to 50%. This cycle-off percentage can be matched to average characteristics in a utility service area. However, since the percentage is preset it cannot be matched to the diversity in natural thermostatic cycling across houses. For example, this study finds that 16% of the houses had duty cycles below 0.50 on the hottest summer days during the hours selected for duty cycling. These houses cannot provide load savings with a DLC controller preset to 50% or less off-time. Fully 60% of the houses had duty cycles below 0.75 for this period, yielding no savings at 25% off-time and half the expected savings at 50% off-time. Two methods are proposed for improving efficiency of DLC programs: pre-screening based on measured duty cycle, and adoption of advanced DLC controllers. For the study sample, pre-screening is found to be not economical if it requires a separate site visit - surprisingly, it is cheaper to install equipment even in residences which will not yield by any load savings than to pay for screening visits. A more satisfactory improvement in DLC program efficiency would result from advanced DLC controllers with redesigned cycling logic. In place of the current fixed cycle-off percentage, an advanced controller is proposed which would choose the best cycle-off time based on each individual house's natural thermostatic duty cycle. The results indicate that this advanced controller would reduce program costs per kW saved by 29%. Such a controller would also achieve load savings more equitably, spreading cooling reductions more evenly across participants, and eliminate current programs' perverse incentive to oversize cooling equipment.