One method of designing non-recursive digital filters to arbitrary frequency specifications is to use optimization. However, the program will only minimize one particular performance index which may not give the ‘best’ filter for the particular application. Furthermore, when performing the optimization the errors at specified points are usually weighted and the weights are generally chosen in a fairly arbitrary manner. This paper describes how two different optimization runs can be used to give initial conditions which enable the final filter characteristic to be simply adjusted, interactively with the computer, thus avoiding the problem of how best to choose these weights.