Program invariants are statements asserting properties of programs at certain points. They can assist developers and testers in understanding the program, and can be used for automated formal verification of the program. However, despite their usefulness they are often omitted from code. Dynamic invariant detection is a technique that discovers program invariants by observing execution of the program. One type of invariants that presents challenge to this technique is conditional invariants, which are considered to be computationally infeasible to be computed exhaustively. We present a new approach to assist conditional invariants detection, by analysing test suites used to drive the execution of the programs for their use of equivalence partitioning – a very common testing technique – and inferring conditional invariants from this information. A prototype implementation, named Yacon, is developed to work in conjunction with a mature dynamic invariant detection tool Daikon. Given a set of splitting conditions, Daikon can use them to infer conditional invariants. Yacon attempts to recover partitioning information from a given test suite, producing splitting conditions as a result. We introduced two strategies to recover partitioning information, one based on the presence of boundary value analysis testing technique; the other based on invariants within the test suite itself. We evaluated the effectiveness of each recovery strategy and the approach as a whole, and found that our approach can help make Daikon perform significantly better. However, the two recovery strategies only work well in limited circumstances, suggesting possible improvement in finding more effective recovery strategies.