In this article we develop a conjugacy theory in quasiconvex analysis, in which no lower semicontinuity or normality assumption is needed to ensure the coincidence of the second conjugate of any function with its quasivonvex hull. This is made by an extension of the concept ofH-conjugation, and is based on a separation theorem by general halfspaces. The theory is applied in mathematical programming to define dual problems, which consist in maximizing a quasiconcave function of matricial variable, the optimum being always attained. The absence of duality gap is equivalent to the quasiconvexity of the perturbation function at the origin. A Lagrangian for general problems is studied and compared with the one of Luenberger in the case of vertical perturbations.