Global longitudinal strain (GLS) at rest on two-dimensional speckle tracking echocardiography (2D STE) was demonstrated to help detect coronary artery disease (CAD). However, the optimal cut-off point of GLS and its diagnostic power for detecting critical CAD in non-diabetes mellitus (DM) patients are unknown. In the present study, 211 patients with suspected CAD were prospectively included, with DM patients excluded. All patients underwent echocardiography and subsequently coronary angiography within 3 days. Left ventricular (LV) GLSs were quantified by 2D STE. Territorial peak systolic longitudinal strains (TLSs) were calculated based on the perfusion territories of the 3-epicardial coronary arteries in a 17-segment LV model. Critical CAD was defined as an area stenosis >= 70% in >= 1 epicardial coronary artery (>= 50% in left main coronary artery). Totally 145 patients were diagnosed as having critical CAD by coronary angiography. Significant differences were observed in all strain parameters between patients with and without critical CAD. The area under the receiver operating charcteristic (ROC) curve (AUC) for GLS in the detection of left main (LM) or threevessel CAD was 0.875 at a cut-off value of -19.05% with sensitivity of 78.1% and specificity of 72.7%, which increased to 0.926 after exclusion of apical segments (cut-off value -18.66%; sensitivity 84.4% and specificity 81.8%). The values of TLSs were significantly lower in regions supplied by stenotic arteries than in those by non-stenotic arteries. The AUC for the TLSs to identify critical stenosis of left circumflex (LCX) artery, left anterior descending (LAD) artery and right coronary artery (RCA), in order of diagnostic accuracy, was 0.818 for LCX, 0.764 for LAD and 0.723 for RCA, respectively. In conclusion, in non-DM patients with suspected CAD, GLS assessed by 2D STE is an excellent predictor for LM or three-vessel CAD with high diagnostic accuracy, and a higher cut-off point than reported before should be used. Excluding apical segments in the calculation of GLS can further improve the predictive accuracy of GLS. It is unsatisfactory for TLSs to be used to identify stenotic coronary arteries.