Abstract: For the linear discrete stochastic descriptor system with multisensors, by using the singular value decomposition, it is transformed into two reduced order coupled subsystems. By the modern time series analysis method, based on the autoregressive moving average (ARMA) innovation model, and white noise estimation theory, a distributed steady-state Kalman fuser with a three-layer fusion structure is presented, which consists of two weighted fusers and two composite fusers. The first layer gives the state fusers of subsystems which realize a decoupled fusion of components for each subsystem. The second layer gives the transformed state fuser which realizes a decoupled fusion between two subsystems. The third layer gives the original state fuser. It can handle the fused filtering, smoothing, and prediction problems in a unified framework. In order to compute the optimal weights, the formulas of computing the cross-covariances among local estimation errors are presented. It is proved that its accuracy is higher than that of each local estimator. A Monte Carlo simulation example shows its effectiveness.