We consider the variance of the reward until absorption in a Markov chain. This variance is usually calculated from the second moment (expectation of the square). We present a direct system of equations for the variance, involving the first moment (expectation) but not the second moment. This method is numerically superior to the calculation from the second moment.