Model Specification Test is a key step in financial modeling to reduce the model risk． Based on the first HJ distance proposed by Hansen and Jagannthan( 1997) ，Taiwan market data are used to test the model specifications of eight linear factor models ( including models based on financial asset prices) ，and the impacts of model specification assumptions on parameter tests are discussed． The paper finds that，under the 5% sig-nificance level，there exists model misspecification problems for all unconditional models and only the condi-tional versions of FF3，LM，VanM and SkewM are acceptable right models． Meanwhile，taking potential mod-el misspecification into account may detect the factors’pricing ability more efficiently． Assuming a model is rightly specified overestimates the t absolute values of SDF parameters，resulting in a“pseudo-pricing”for some parts of the factors．