Background: 3 T MRI has been adopted by some centers as the primary choice for assessment of myocardial perfusion over conventional 1.5 T MRI. However, there is no data published on the potential additional value of incorporating semi-quantitative data from 3 T MRI. This study sought to determine the performance of qualitative 3 T stress magnetic resonance myocardial perfusion imaging (3 T-MRMPI) and the potential incremental benefit of using a semi-quantitative perfusion technique in patients with suspected coronary artery disease (CAD). Methods: Fifty eight patients (41 men; mean age: 59 years) referred for elective diagnostic angiography underwent stress 3 T MRMPI with a 32-channel cardiac receiver coil. The MR protocol included gadolinium-enhanced stress first-pass perfusion (0.56 mg/kg, dipyridamole), rest perfusion, and delayed enhancement (DE). Visual analysis was performed in two steps. Ischemia was defined as a territory with perfusion defect at stress study but no DE or a territory with DE but additional peri-infarcted perfusion defect at stress study. Semi-quantitative analysis was calculated by using the upslope of the signal intensity-time curve during the first pass of contrast medium during dipyridamole stress and at rest. ROC analysis was used to determine the MPRI threshold that maximized sensitivity. Quantitative coronary angiography served as the reference standard with significant stenosis defined as >70 % diameter stenosis. Diagnostic performance was determined on a per-patient and per-vessel basis. Results: Qualitative assessment had an overall sensitivity and specificity for detecting significant stenoses of 77 % and 80 %, respectively. By adding MPRI analysis, in cases with negative qualitative assessment, the overall sensitivity increased to 83 %. The impact of MPRI differed depending on the territory; with the sensitivity for detection of left circumflex (LCx) stenosis improving the most after semi-quantification analysis, (66 % versus 83 %). Conclusions: Pure qualitative assessment of 3 T MRI had acceptable performance in detecting severe CAD. There is no overall benefit of incorporating semi-quantitative data; however a higher sensitivity can be obtained by adding MPRI, especially in the detection of LCx lesions.