To assess a novel approach for automatic flow velocity computation in deriving quantitative flow ratio (QFR) from coronary angiography. QFR is a novel approach for assessment of functional significance of coronary artery stenosis without using pressure wire and induced hyperemia. Patient-specific coronary flow is estimated semi-automatically by frame count method, which is subjective and inconvenient in the workflow of QFR analysis. The vascular structures were automatically delineated from coronary angiogram. Subsequently, the centerline of the interrogated vessel was extracted from the delineated lumen on each image frame and the change in the length of centerline was used to compute the flow velocity, which provided patient-specific flow for computation of QFR (QFR_auto). A parameter derived from the increase in centerline length was used to automatically quantify the stability of contrast flow. From the two angiographic image runs used for three-dimensional angiographic reconstruction, the one with better stability was used to compute QFR_auto. QFR_auto was assessed in all patients enrolled in the FAVOR II China study, and compared with the commercialized QFR computational method based on frame count (QFR_count), using pressure wire-based fractional flow reserve (FFR) as the reference standard. Out of 328 vessels with paired FFR data, QFR_auto was successfully computed on 325 (99%) vessels with acceptable stability in filling of contrast flow. The flow velocity computed by the proposed approach had a weak to moderate correlation with the frame count method (r = 0.37, p < 0.001), with mean differences of − 0.02 ± 0.07 m/s (p < 0.001). QFR_auto had good correlation (r = 0.96, p < 0.001) and agreement (mean difference: − 0.01 ± 0.04, p < 0.001) with QFR_count. Good correlation (r = 0.83, p < 0.001) and agreement (mean difference: 0.01 ± 0.06, p = 0.016) were also observed between QFR_auto and FFR. Using FFR ≤ 0.80 to define functional significance of coronary stenosis, the overall diagnostic accuracy for QFR_auto was 93.2% (95% CI 90.5–96.0%). The area under the receiver-operating characteristic curve did not differ significantly between QFR_count and QFR_auto (difference: 0.00; 95% CI − 0.01 to 0.01; p = 0.529). Sensitivity, specificity, positive likelihood ratio, and negative likelihood ratio for QFR_auto were 92.4% (95% CI 86.0–96.5%), 93.7% (95% CI 89.5–96.6%), 14.7 (95% CI 8.7–25.0), and 0.1 (95% CI 0.0–0.2), respectively. Automatic computation of patient-specific coronary flow velocity based on coronary angiography is feasible. Assessment of QFR based on this novel approach had good diagnostic accuracy in determining the functional significance of coronary stenosis.