▼ We consider the inference problem of a finite mixture model based on data from multiple samples, each of which is from a mixture of two common components. Under the assumption that the ratio of the two component densities takes a known parametric form, we obtain maximum semiparametric likelihood estimates of the parameters via EM or MM algorithms, and establish the large sample results for those estimators. We then develop empirical likelihood ratio-based statistics for constructing confidence intervals for and testing statistical hypotheses on mixing proportions. We show that the statistics are asymptotically chi-square distributed. In addition, a goodness-of-fit test is proposed for testing the density ratio assumption. Simulation studies are carried out to evaluate the performances of the proposed statistics and tests. Advisors/Committee Members: Zhang, Biao.