Compositional data consist of vectors of nonnegative values summing to a unit. In this paper the PLS path modeling is applied to the compositional data after logratio transformation and a method of multiple linear regression modeling is put forward when the dependent and independent variables are all correlated compositional data. For compositional data this modeling method can satisfy the unit-sum constraint and eliminate the harms that derive from the complete multicollinearity. By the case of a regression model built on Beijing＇ s three industries structure data of investment, GDP and employment, it is proved that the modeling can clarity the property that contained in the compositional data effectively