In survey sampling, generally the use of auxiliary information at the estimation stage enriches the efficiency of estimators. Product estimator suggested by Murthy (1964) is more efficient than mean per unit estimator when there exists a negative correlation between study variable and auxiliary variable. In surveys, whenever the population mean of the auxiliary variable is not known in advance, one can use two- phase sampling scheme or double sampling scheme. In this paper an attempt has been made to develop three product type estimators to estimate finite population mean using two phase sampling schemes suggested by Bose (1943) First estimator is constructed using known coefficient variation of study variable and other two estimators are constructed using estimated coefficient variation of study variable. The efficiencies are compared with mean per unit estimator, conventional two-phase product estimator and two-phase product type exponential estimator suggested by Singh and Vishwakarma (2007), both theoretically and empirically.