A system and program product for automatic identification of subroutines from test scripts is disclosed. An instruction class may be coded as a subroutine vector in a vector space model. A test script action may be coded as a test script vector in a vector space model. The test script vector may be compared to the subroutine vector. The test script vector may be identified and labeled. The test script vector may be labeled as a new test script action if the result of comparing is below a threshold value. An identifier may be associated with a new test script action.