In this paper we address the problem of the non-LTE (NLTE) inversion of line profiles by means of a nonlinear least-squares minimization procedure combined with very efficient multilevel transfer methods. Our approach is based on the concept of response functions, which measure the first-order response of the emergent profiles to changes in the atmospheric conditions. We introduce the fixed departure coefficients (FDC) approximation in order to compute these response functions in a fast and straightforward manner. The accuracy of this approximation is checked comparing FDC response functions with those obtained from full NLTE computations. An NLTE inversion code based on these response functions has been developed and extensively tested. Reference synthetic profiles, similar to those expected from real observations, are given as input to the inversion algorithm and the recovered models are shown to be compatible with the reference models within the error bars. Our NLTE inversion code thus provides a new tool for the investigation of the chromospheres of the Sun and other stars.