In one embodiment, a method for troubleshooting a fault to determine a root cause of the fault. A Bayesian network model is created based on information obtained from a Fault Isolation Manual (FIM), where the FIM provides tests to be performed in troubleshooting the fault to determine a root cause of the fault. Heuristics are used to determine a structure and conditional probabilities for the Bayesian network. A plurality of test costs inherent in the FIM are imputed by first generating a plurality of constraints between the cost of each test and fault probabilities that hold for all fault scenarios. A linear programming algorithm is used to solve the plurality of constraints, and to construct a tuned Bayesian network model. The tuned Bayesian network model is used to iteratively rank likely faults according to their probabilities given accumulating test evidence, and to rank pending tests according to their value.