In this paper, a target detection technique based on the fusion of multipath signals in a Bayesian framework is proposed with application to the HF Over-The-Horizon-Radar (OTHR). The detection consists of a generalized likelihood ratio test and a data fusion process, where target measurement likelihood is derived based on a set of multi-path models which are characterized by a known Doppler transition with additive Gaussian noise. The latter describes the relationship between the Doppler frequencies propagated along each of the multiple paths. A centralized Bayesian fusion is applied to compute an overall amplitude of the detection iteratively, where the posterior probability density of Doppler frequency for a specific propagation path from the underlying target is estimated via a sequential Bayesian estimator. Compared to the conventional detection method used in OTHR, the proposed detection technique is able to increase detection SNR by fusing target detections over all paths. Results from our numerical simulations are presented, which show the effectiveness of the proposed algorithm.