Three different deconvolution techniques for quantifying cerebral blood flow (CBF) from whole brain T*(2)-weighted bolus tracking images were implemented (parametric Fourier transform P-FT, parametric single value decomposition P-SVD and nonparametric single value decomposition NP-SVD). The techniques were tested on 206 regions from 38 hyperacute stroke patients. In the P-FT and P-SVD techniques, the tissue and arterial concentration time curves were fit to a gamma variate function and the resulting CBF values correlated very well (CBF(P-FT) = 1.02 x CBF(P-SVD), r(2) = 0.96). The NP-SVD CBF values (i.e., original unfitted curves were used) correlated well with the P-FT CBF values only when a sufficient number of time series volumes were acquired to minimize tracer time curve truncation (CBF(P-FT) x 0.92 x CBF(NP-SVD), r(2) = 0.88). The correlation between the fitted CBV and the unfitted CBV values was also maximized in regions with minimal tracer time curve truncation (CBV(fit) = 1.00 x CBV(unfit), r(2) = 0.89). When a sufficient number of time series volumes could not be acquired (due to scanner limitations) to avoid tracer time curve truncation, the P-FT and P-SVD techniques gave more reliable estimates of CBF than the NP-SVD technique.