Utilizing the concept of unbiased estimating functions combined with the concept of $\phi$-divergence of Csiszár (1963), we introduce the method of minimum empirical $\phi$-divergence estimation and inference, which can be thought of as a generalization of empirical likelihood method of Owen (1988) and Qin and Lawless (1994). Efficiency results for estimators are obtained. Given the parallels with the conventional likelihood, classical-type tests based on the empirical $\phi$-divergence for a simple parametric hypothesis and the moment conditions are constructed.