A UML-driven technique for detection of performance bottlenecks in concurrent real-time systems is presented. The approach is based on comprehensive analysis of control flow in two types of UML 2.x behavioral models: sequence diagrams and interaction overview diagrams. The technique takes an input the runtime durations of tasks and uses the Program Evaluation and Review Technique (PERT) to pinpoint performance bottlenecks in UML-based control flow information of a concurrent real-time system. Since design UML models are usually developed and are available already for most object-oriented systems, the technique prevents the need to construct specific-purpose performance models such as Layered Queuing Networks. Application of the technique on an example control software system demonstrates the applicability and effectiveness of the technique in pinpointing performance bottlenecks.