High-accuracy position awareness is essential to many applications, such as indoor navigation and autonomous vehicles. Wireless localization is a promising positioning service provider with the merits of extensive coverage and low cost but degrades in complex multipath propagation environments. This paper proposes a tensor-based algorithmic framework for localization in multipath environments by exploiting sparse spatio-temporal features of the received waveforms. Specifically, we construct low-rank tensors to characterize sparse spatio-temporal features, which can separate coherent multipath signals hinging on the uniqueness of tensor decomposition. Compared with related tensor-based works, our method does not rely on array configurations or signal structures, revealing its potential for broad use in multipath estimation. Position-related parameters are further extracted from tensor decomposition results, where a method based on the Chinese remainder theorem (CRT) is developed to retrieve distance information from the carrier part. Simulation results show that our method yields high-accuracy localization performance in complex multipath environments.