Accurate prepayment models are critical to the valuation of MBS (mortgage backed securities). They also play an important role in the efficient hedging of interest rate risk associated with mortgage funding. We have developed prepayment models using logistic regression for Dutch mortgage loans. We distinguish two types of prepayment: economical prepayment (prepayment without selling the house) and social prepayment (prepayment with selling the house). Prepayment caused by default is included into social prepayment. Using more than 100,000 loan parts and more than 3 million data observations between March 1999 and December 2005, the prepayment probabilities are estimated on an individual loan part level. In the chosen data period, the interest rate dropped to a historical low level not seen since 1960. The models have been validated by backtesting them against historical CPR (constant prepayment rate). The prepayment behaviour in the Netherlands shows unique characteristics comparing to other countries due to the existence of prepayment penalty and tax deduction on penalty interest in the Netherlands. In addition to refinancing incentive, market interest rate, LTV (loan to value), age of borrower, seasoning, seasonality and property type, we found that months-in-arrear and time-on-job are important predictors for prepayment. Furthermore, variables influence the two types of prepayment in distinctive ways and have different relative importance for the two types of prepayments. A previous reported December effect is found to be only relevant to economical prepayment while the time-on-job variable is found to be only relevant to social prepayment.