The invention discloses a novel multi-source satellite altimeter fusion method. The method comprises the steps of decomposing a background error covariance matrix B = sigma C sigma by adopting a diagonal matrix and a correlation coefficient matrix based on an optimal algorithm of two-dimensional variation, estimating and constructing background error correlation by utilizing a Gaussian function, fitting Gaussian distribution through a least square method, obtaining background error correlation coefficient scale, adopting a multi-time satellite altimeter to observe the density to determine the fusion duration, and fusing observation data to obtain a fusion product. According to the method, a two-dimensional variational method is taken as a fusion basis, evolution errors of same-orbit observation are considered, background error covariance correlation coefficient scales are selected according to the altimeter observation density, a more reasonable background error correlation coefficient representation function is constructed, changes of background error correlation coefficient scales of different areas are considered, and the fusion field of the previous day is used as the background field, so that more small-scale signals are reserved in the fusion product, and the effective resolution of the fusion product is remarkably improved.