This paper critically reviews current practices for measuring credit r
isks of derivative instruments. It argues that there are two major pro
blems with the standard measurement approach. First, it uses models of
the stochastic behavior of financial variables while ignoring both th
eir inherent oversimplification and the uncertainty in their parameter
s. Second, it ignores the correlations among exposures on derivative i
nstruments and the probabilities of counterparty default. This paper d
emonstrates that these practices can produce large errors in the estim
ation of distributions of both future credit exposures and future cred
it losses.