Abstract Cooking tests were conducted in randomly selected school kitchens in the Sauri Millennium Villages Project site, located in Siaya District of Nyanza Province in Western Kenya. The tests compared fuel consumption measurements obtained using a traditional three-stone fire with those from newly introduced institutional stoves based on the “rocket” design. The key metric used was Specific Fuel Consumption (SFC), defined as the weight of firewood consumed in cooking a single batch of food divided by the total weight of food, measured after cooking. Tests followed the normal cooking practices in the school kitchens and included the typical range of foods prepared for midday school meals programs. The study included two types of tests: paired tests, in which most conditions were controlled between one test conducted on a three-stone fire and a matching test conducted on a “rocket” stove; and unpaired tests, in which conditions were similar, but not strictly controlled, among two large sets of relatively independent three-stone fire and rocket stove tests. Results from both paired and unpaired experiments, averaged across all types of food cooked, showed that the use of rocket stoves yielded significantly lower SFC values without prolonging cooking time when compared with three-stone fires. An analysis comparing results from paired and unpaired cooking tests suggests that, due to high variance and sources of bias in unpaired tests, experimental design should favor paired tests.