Training with limited carbohydrate availability can stimulate adaptations in muscle cells to facilitate energy production via fat oxidation. Here we investigated the effect of consistent training in the fasted state, vs. training in the fed state, on muscle metabolism and substrate selection during fasted exercise. Twenty young male volunteers participated in a 6-wk endurance training program (1-1.5 h cycling at ∼70% Vo(₂max), 4 days/wk) while receiving isocaloric carbohydrate-rich diets. Half of the subjects trained in the fasted state (F; n = 10), while the others ingested ample carbohydrates before (∼160 g) and during (1 g·kg body wt⁻¹·h⁻¹) the training sessions (CHO; n = 10). The training similarly increased Vo(₂max) (+9%) and performance in a 60-min simulated time trial (+8%) in both groups (P < 0.01). Metabolic measurements were made during a 2-h constant-load exercise bout in the fasted state at ∼65% pretraining Vo(₂max). In F, exercise-induced intramyocellular lipid (IMCL) breakdown was enhanced in type I fibers (P < 0.05) and tended to be increased in type IIa fibers (P = 0.07). Training did not affect IMCL breakdown in CHO. In addition, F (+21%) increased the exercise intensity corresponding to the maximal rate of fat oxidation more than did CHO (+6%) (P < 0.05). Furthermore, maximal citrate synthase (+47%) and β-hydroxyacyl coenzyme A dehydrogenase (+34%) activity was significantly upregulated in F (P < 0.05) but not in CHO. Also, only F prevented the development exercise-induced drop in blood glucose concentration (P < 0.05). In conclusion, F is more effective than CHO to increase muscular oxidative capacity and at the same time enhances exercise-induced net IMCL degradation. In addition, F but not CHO prevented drop of blood glucose concentration during fasting exercise.