Estimates of the trade diversion (TD) effect of preferential trade agreements (PTAs) tend to be much smaller than that of the trade creation (TC) effect. This paper examines two sources of estimation bias of the TD effect. The first bias arises from the failure to recognise that the concept of TD is inapplicable for a substantial proportion of PTAs, and thus, previous studies wrongly count cases where trade cannot be diverted as cases where trade is not diverted. The second bias arises from the difficulty in controlling for multilateral resistance and other unobserved time-varying country heterogeneity when estimating both TC and TD effects simultaneously due to perfect multicollinearity. This paper corrects these two biases using a new measure of TD which provides a better mapping between the concept and data characteristics. Removing the two sources of bias leads to a multifold increase in the estimate of the TD effect. It is found that the total TD effect is comparable to the TC effect in dollar terms.