We evaluated the optimal flow velocity of transcranial doppler (TCD) in detecting unilateral middle cerebral artery (MCA) stenosis and stenosis grading by magnetic resonance angiography (MRA) as the reference standard. 302 nonconsecutive patients with unilateral MCA stenosis detected by TCD underwent MRA of the intracranial arteries. The peak systolic velocity (PSV), mean flow velocity (MFV), and end-diastolic velocity (EDV) of each MCA were recorded. 604 MCA were categorized into four groups depending on the stenosis severity: normal MCA (n = 319, 52.8%), mild stenosis (n = 94, 15.6%), moderate stenosis (n = 66, 10.9%), and severe stenosis (n = 125, 20.7%). Significant differences in PSV, MFV, and EDV between these four groups were observed (P < 0.001, respectively). The optimal cutoff velocities for detecting MCA stenosis were: PSV = 160 cm/s, MFV = 100 cm/s, EDV = 60 cm/s; the optimal cutoff points to distinguish mild from moderate stenosis were: PSV = 200 cm/s, MFV = 120 cm/s, EDV = 80 cm/s; the cutoffs to distinguish moderate from severe stenosis were: PSV = 280 cm/s, MFV = 180 cm/s, EDV = 110 cm/s. Using PSV as the diagnostic criteria, the correlation for diagnosing MCA stenosis using TCD and MCA was good (Kappa number κ = 0.668); using as MFV criteria, κ = 0.641. The optimal cutoff PSV values in stenosis grading on TCD were 160, 200, and 280 cm/s. The optimal cutoff MFV values were 100, 120, and 180 cm/s. PSV is more accurate than MFV in detecting and grading MCA stenosis.