In survey sampling, stratification is helpful in improving precision of estimators over simple random sampling in case of heterogeneous population. In the present paper, a difference-cum exponential type estimator of population mean under two phase stratified random sampling design has been proposed for the case of heterogeneous population. The expressions for bias and mean squared error of the proposed estimator have been obtained up to first order of approximation. It has been shown that proposed estimator is efficient than linear regression estimator under the same sampling design for some range of variation in the values of constants involved. The results obtained have also been illustrated numerically as well as graphically by taking data from the population considered in the literature