The focal point of this paper is a new result on the probabilistic robustness of a stochastic first order filter. For a first order filter transfer function, G(h, r ) , we allow a class of probability distributions F for the time constant T and consider the following question: Given frequency iu' > 0 and unknown probability distribution .f E F, to what extent can the expected filter gain !/(d. T ) = r)l deviate from some desired nominal value, T,,)? It tums out that the deviations of concem are surprisingly low. For example, with 20% variation in T , the expected filter gain deviates from g ( d . T O ) by no more than U.4% of the zero frequency gain. In addition to performance bounds such as this, we also provide a so-called universalfigure of merit. The word "universal" is used because the performance bound attained holds independently of the nominal ro, the frequency w 2 0 and the admissible probability distributions .f E F.