Conventional blood flow velocity measurement using ultrasound is capable of resolving the axial component (i.e., that aligned with the ultrasound propagation direction) of the blood flow velocity vector. However, these Doppler-based methods are incapable of detecting blood flow in the direction normal to the ultrasound beam. An algorithm which measures the lateral blood flow velocity using speckle size change with scan velocity was developed in our previous studies. This method uses the apparent speckle size change that occurs when scatterers are moving relative to the spatial rate of A-line acquisition. Our previous results showed that the estimation error of this algorithm increases with increasing flow gradient and random scatterer movement. In this paper, the relationship between the estimation performance and flow gradient, random scatterer movement and ROI size is investigated and quantitatively assessed. Simulated blood flow data with and without flow gradient and random scatterer movement were generated by the Field II simulation program. The flow gradient is introduced by a parabolic flow profile in the simulated vessel and the random scatterer movement is generated by adding Gaussian noise to the scatterers' position with a standard deviation as much as one-tenth of the speckle cell size in each direction. Our results showed that: 1) in plug flow, estimation error decreases with increasing ROI size, with an average minimum error below 5%. An optimal ROI size exists in both directions, which is 2.5 axial speckle cell lengths axially and 30 lateral speckle cell widths laterally; 2) the estimation error increases up to 10% with flow gradient; 3) an optimal lateral ROI size still exists given the presence of a flow gradient; 4) the estimation error increases with increasing axial ROI size since the correlation length shortens by the introduction of a flow gradient; 5) in addition to the previous results, when random scatterer movement is introduced into the blood flow, the average estimation error is worse by about a factor of three than data without random scatterer movement.