MR perfusion imaging can be used to estimate parameters indicating the metabolic process, including cerebral blood flow (CBF). Specifically singular value decomposition (SVD) method is applied in this paper to obtain CBF with a predefined conventional arterial input function (AIF), and the effects of threshold and tracer delay are investigated. While the simulation results show that SVD method can estimate CBF with good accuracy by using different thresholds for different flow values, the method is found to be sensitive to tracer delay. A delay correction scheme is advocated, where CBF is determined by SVD after time-shifting of the tracer concentration curve. With the correction the simulation results are considerably improved in the flow estimation.