We analyse job-training effects on Korean women for the period January 1999 to March 2000, using a large data set of size about 52,000. We employ a number of estimation techniques: Weibull MLE and accelerated failure time approach, which are both parametric; Cox partial likelihood estimator, which is semiparametric; and two pair-matching estimators, which are in essence nonparametric. All of these methods gave the common conclusion that job training for Korean women increased their unemployment duration. The trainings were not cost-effective in the sense that they took too much time ‘locking in’ the trainees during the training span, compared with the time they took to place the trainees afterwards. Despite this negative finding, some sub-groups had positive effects: white-collar workers trained for finance/insurance or information/communication. Copyright © 2005 John Wiley & Sons, Ltd.