An order estimation of a linear time invariant system has been developed. The order estimation (the number of sets of significant system parameters) is based on function elimination filters. In presence of noise, the estimation procedure has been improved by using the composite hypothesis and the maximum likelihood ratio. Finally, the estimated system order can be used to estimate the system parameters. It is to be assumed that the investigated system is not overdumped, has no aliasing problems and the system input is "white noise". The proposed method differs from other estimation methods because the system order can be found without knowledge about the system parameters.