Soil moisture is an integral part of the water, energy and biogeochemical cycle. The information about soil moisture is of great significance to researches on water resources management, agricultural production and climate change. Soil moisture monitoring can be divided into three categories in light of data acquisition method: direct measurement at monitoring sites, simulation and assimilation of soil moisture, and inversion based on remote-sensing data. The remote sensing technology features largescale synchronous observation, covering a range that is not limited by the distribution of ground stations. Then the remote-sensing data based inversion algorithm of soil moisture is an important means of obtaining soil moisture information. However, as soil moisture is strongly influenced by a variety of factors, such as soil properties, surface coverage and meteorological conditions, it is high in spatial heterogeneity. So, it is very difficult to derive large-scales high quality soil moisture data just based on inversion with a single method or single data source. In this paper, factors affecting the inversion of soil moisture were collated, four synthetic multi-featured models for soil moisture inversion were summarized, and existing problems and developmental trends of the inversion processes analyzed. The eigenvariances currently used in soil moisture inversion can be generally sorted into three categories: soil, vegetation and meteorological characteristics. Soil characteristics can be further divided into soil optical reflectance, thermal infrared, microwave brightness and temperature and microwave backlash scattering coefficient, and vegetation characteristics into vegetation optical reflectance and thermal infrared, while meteorological characteristics include rainfall, wind speed, and evapotranspiration and so on. In this paper, synthetic models for multfeatured eigenvariance inversion of soil moisture were summarized, that is, Temperature Vegetation Soil Moisture Dryness Index model (TVMDI), partition statistics model, benchmark value plus variation model, and artificial neural network model. TVMDI is a cubic model based on land surface temperature, vertical vegetation index and soil moisture, and its use enhances correlativity of prediction with measured value. The partition statistics model is to choose an optimal model for each region for inversion, through analyzing types of land cover. The benchmark value plus variation model is to divide the variation of soil moisture during a specified observation period into benchmark value and variation. The former represents the bottom of soil moisture during that period, and the latter depends on precipitation, evapotranspiration and some other meteorological factors, while integrating remote-sensing meteorological information. The artificial neural network model integrate multi-featured eigenvarianes into soil moisture inversion. The analysis of existing problems in and developmental trend of the use of eigenvariance in soil moisture inversion process indicates that the research on adoption of the theory of eigenvariance in soil moisture inversion is insufficient and comprehensive application of the theory is not deep enough, and stresses that coupled application of various ergenvariances may improve accuracy of soil moisture inversion, which is the hot spot of future researches.