This paper describes a method that combines Kalman-filter and neural network to form an efficient data fusion technique for estimating payload in the bucket of a moving loader. Kalman-filter is used to find the signal levels from noisy measurement data before the data is fed to the neural network. Neural network is then used to form the nonlinear connection between the indirect measurements describing the load and the actual load in the bucket. The results show that the used combination of these different methods offers a viable solution for estimating the payload.