Abstract: This paper is concerned with the minimal order filter for a singular linear time-invariant continuous stochastic system. The equations are derived for the reduced order（n-m+r order） linear optimal filter without knowing statistical property of dynamic noise, under the assumption that some measurements are noise-free and that there is a finite number of derivatives for the exact measurements. By means of using the simplest matrix generalized inverse,a matrix L is chosen in order to eliminate the impact of dynamic noise. When the observational equation is under the hypothesis on singularity and a solution of the lemma exists,we show that there exists a minimal order filter. The possibility of the reduced order filtering method is illustrated by a simple example at the end.