This paper addresses the problem of targeted test generation for actor systems. Specifically, we propose an approach to generate system-level tests which cover a given code location in an actor system. The approach consists of two phases. First, static analysis is used to construct a message flow graph (MFG). The MFG is an abstraction of the entire actor system that captures potential actor interactions defined by a program. Second, a backwards symbolic execution (BSE) from a target location to an ’entry point’ of the actor system is performed. The BSE uses the MFG constructed in the first phase to guide execution across actors. Because a BSE generates a huge search space of potential entry points, heuristics combined with a feedback-directed technique are used to guide and prune the search. We implement our approach in a tool called Tap for Java Akka programs, and evaluate Tap on the Savina benchmarks and four open source projects. Our evaluation shows that the Tap achieves a relatively high target coverage (78% on 1,000 targets) and detects six previously unreported bugs in the subjects.