This document discusses using iterative refinement with extra precision to solve linear systems Ax = b more accurately. It proposes computing residuals and updates with higher precision than the working precision to reduce errors. Test results on random systems show this approach reduces both backward and forward errors much further than working precision alone. The costs of extra precision and iterative refinement are analyzed. In most cases, little extra precision and a few refinement iterations are needed to reduce errors to the limit of the higher precision.