The Wiener model is a block oriented model where a linear dynamic system block is followed by a static nonlinearity block. The dominant method to estimate these components has been to minimize the error between the simulated and the measured outputs. This is known to lead to biased estimates if disturbances other than measurement noise are present. For the case of more general disturbances we present Maximum Likelihood ex- pressions and provide algorithms for maximising them. This includes the case where disturbances may be coloured and as a consequence we can handle blind estimation of Wiener models. This case is accommodated by using the Expectation-Maximisation algorithm in combination with parti- cles methods. Comparisons between the new algorithms and the dominant approach conrm that the new method is unbiased and also has superior accuracy.