The possibility that the effect of monetary policy on output may depend on whether credit conditions are tight or loose can be expressed as a non-linearity in the relation between real money supply and output, of which a simple case is a threshold effect. In this case, consistent with the credit-rationing model of Blinder (1987), the monetary variable has a more powerful effect if it is below some threshold than when it is above. Testing for the importance of this threshold is straightforward if the appropriate threshold value is known a priori, but where the value is not known and must be chosen based on the sample, the testing problem becomes more difficult. We apply recently-developed tests applicable in this situation to both US and Canadian data, and find substantial evidence of a threshold effect, particularly in US data. However, the estimated threshold values are high.