The purpose of this study was to evaluate different efficiency indices, i.e., gross (GE: no baseline correction), net (NE: resting metabolism as baseline correction), and work (WE: unloaded exercise as baseline correction), to reveal the effect of endurance training on mechanical efficiency. Nine healthy sedentary women undertook an incremental test and submaximal cycling exercise, at an intensity corresponding to 50% of the pretraining peak oxygen uptake, before and after 6 weeks of endurance training (18 sessions of 45 min). The training effects on efficiency indices were tested by comparisons based on GE, NE, and WE as well as by the differences between the percentage changes of all indices (% GE, % NE, % WE). Endurance training resulted in significantly higher GE (+ 11.1%; p < 0.001) and NE (+ 9.1%; P < 0.01). Only minor significant improvement (+ 2.4%; p < 0.05) was observed with the WE index because the value used for baseline subtraction was significantly reduced by the training sessions, due perhaps to improvement in pedaling skill. As a consequence, % WE was significantly lower than % GE (p < 0.01) and % NE (p < 0.05), while % GE and % NE were not significantly different. We conclude that mechanical efficiency of cycling increases with training in women previously unfamiliar with cycling, and that the WE index is less sensitive to this training effect than GE and NE indices. Key words: gross efficiency, net efficiency, work efficiency, internal work, cycle ergometer