This article derives improvements to the LSMC for derivatives pricing. For single asset pricing, the results imply that our method can even raise computational speed by 10% to 14% more than that of Choi and Song (2008). We also extend our model for multi-asset pricing and make a comparison with Anderson and Broadie’s (2004) approach in valuing rainbow options. The benchmark values can be exactly covered in 95% confidence intervals and the computational time is reduced by about 94%. Further, we use the Sobol’ sequence to reduce 99% computational time under 0.04%~0.9% pricing errors.