The paper is devoted to the trust assessment problem for specific types of software/hardware systems, namely Systems as Services. We assume that such systems are designed and utilized in all application domains, and therefore the aspects of trust are becoming crucial. Moreover, these systems are mainly used on-demand and are often represented by a composition of 'smaller' services. Thus, an effective method for estimating/assessing the trust level of a given component service (or a system as a whole) needs to be utilized. Most known methods and techniques for trust evaluation mainly rely on the passive testing and system monitoring; in this paper, we propose a novel approach for this problem taking advantage of active testing techniques. Test sequences to be applied to a system/service under test are derived based on determining the critical values of non-functional service parameters. A set of these parameters can be obtained via a static code analysis of the system/service or by add ressing available experts. Machine learning techniques can be applied later on, for determining critical parameter values and thus, deriving corresponding test sequences. The paper contains an illustrative example of RESTFul web service which components are checked w.r.t. critical trust properties