In this paper fully centralized, hierarchical decentralized, and fully decentralized multisensor structures of Kalman filter are introduced and compared with together and with one sensor structure of Kalman filter. Decentralized structures are obtained through parallelism in centralized Kalman filter and give high processing speed. Fully decentralized structure requires no central processor and each processor estimates system's states. Therefore, the system is very resilient to loss of one or more of its processors and its survivability and robustness are increased. Decentralized algorithms are modified so that t-he computation complexity is reduced with keeping the same accuracy. Simulation results 'for time-variant system illustrate clearly that despite of reducing the computation complexity, the accuracy of the estimation is remained the same. Also, multisensory structures have high accuracy of estimation in comparison to one sensor structure.