This paper tests the standard quadratic approximation to central bank preferences on data from Australia and New Zealand, two of the earliest explicit inflation targeting countries. The standard linear-quadratic monetary policy model assumes central bank preferences over key macroeconomic variables, such as inflation and output, can be usefully approximated by a quadratic function. This approximation implies that a deviation from a target is considered to be equally costly irrespective of whether the deviation is positive or negative. Combined with a linear model of the economy, quadratic preferences are useful because they yield a first order condition that implies a linear interest rate reaction function. This paper relaxes the assumption of quadratic preferences by allowing central banks to regard the costs associated with positive and negative output gaps differently. Our models also test for the possibility that positive and negative deviations of inflation from target to be treated differently. During the inflation targeting period in both countries, evidence suggests that we cannot reject quadratic preferences over inflation deviations (from target). We cannot reject that New Zealand's preferences regarding deviations of output from trend are quadratic, but Australia's behaviour does not appear to be consistent with quadratic preferences. Instead, the preferences of the Reserve Bank of Australia appear to be more accurately modelled with an asymmetric loss such that the Reserve Bank of Australia views negative output gaps as more costly than positive output gaps.