In this paper, we estimate a multi-spell duration model of timing of marriage and timing and spacing of children by the Malaysian women. We compute the Heckman-Singer maximum likelihood estimates that control for unobserved heterogeneity nonparametrically, maximum likelihood estimates without correcting for heterogeneity and Cox’s partial likelihood estimates. We find that parameter estimates are very sensitive to the estimation procedure. We use the goodness-of-fit test and Hausman type specification test to choose the appropriate estimates to draw inference about the old-age security hypothesis, replacement effect and sodsex preference hypothesis for the Malaysian families. We find strong evidence for old-age security hypothesis and replacement effect and weak evidence for the son preference hypothesis.