Background Increase in maximal tolerance during high-intensity exercise, above the minimal clinically important difference, reaches only ~45-50% of the subjects with chronic obstructive lung disease (COPD), under pressure support ventilation (NIV). Objectives To characterize a predictive model in COPD for responders (?R?, >33% increase in exercise tolerance) compared to non-responders (?NR?), in a crossover design. Methods We performed cardiopulmonary tests, lung function (including maximal inspiratory pressure before and after maximal exercise, ?MIP), emphysema score, blood gas analysis, and surface electrical activation of the sternocleidomastoid during the inspiratory capacity maneuver (EMGs). A logistic regression modeling was performed to predict binary response. Results Fifteen subjects were categorized as ?R?, and fifteen as ?NR?. At baseline (average±SD), they presented similar age (64±8/64±8 yrs, p=0.920), sex (7/8 vs 11/4, p=0.140), body mass index (27±5/25±6 Kg*m2, p=0.310), FEV1 % predicted (47±13/50±13, p=0.570), monoxide carbon diffusion capacity (51±17/ 52±21, p=0.790), and peak oxygen uptake (75±15/71±22 % predicted, p=0.520), respectively. Blood gas analysis and emphysema score were similar (p>0.05). The only two variables that differ between the groups were the ?MIP (-7±16/+3±10 %, p=0.010) and pre-exercise EMGs (114±57/184±76 ?v, p<0.010), respectively. Thus, the model attained a sensitivity and specificity for correct classification of ?R? of 80% and 87%, respectively. Conclusions Respiratory muscle fatigue after the exercise and weaker pre-exercise EMGs were important predictors of NIV response. Validation of the model is ongoing.