Social assistance programmes (SAPs), understood as non-contributory transfers aimed at ad-dressing poverty, have spread in developing countries since the late 1990s. National govern-ments in Latin America have sought to extend the coverage of SAPs through human devel-opment conditional cash transfer programmes (CCTs). CCTs share several implementation features. First, they employ targeting and selection methods based on means, and proxy means, tests. Research on targeting and selection methods has evolved hand in hand with the adoption of CCTs in Latin America, Africa and South East Asia. Second, CCTs involve the provision of cash transfers directly to households, but with conditions attached to human development objectives. Transfers are given to households in poverty contingent on investment in the human capital formation of their children. A third feature relates to the presence of programme exit conditions. To date, scarce research is available on the design and outcomes associated with exit condi-tions from CCTs. This thesis thus contributes to the literature in the implementation of SAPs by providing a critical examination of exit conditions in SAPs with specific emphasis on CCTs. The thesis provides a systematic theoretical and empirical analysis of the role of exit conditions in the implementation of CCTs. The thesis develops and tests two basic principles underlying the role of exit conditions. First, the exhausted-effectiveness principle suggests that the effectiveness of a CCT varies over time. The research reported in this examines the effectiveness of programme over time with the aim of identifying potential thresholds after which a given SAP's effectiveness de-clines. A two-period child human capital investment model is developed to study analytically the conditions in which programme effectiveness varies over time. This is examined empirically in order to demonstrate the existence of the time-varying effectiveness associated with the implementation of the Colombia's CCT, Familias en Accion. A continuous treatment effect model is estimated following Hirano and Imbens (2004), in which the length of exposure allows for the graphical analysis of dose-response functions. The results indicate that the design of SAPs must take account of time-varying effectiveness. Second, a principle of the non-recurrence of poverty states that beneficiaries should be able to exit an effective programme when two conditions apply: (i) they are not in poverty; and (ii) they face a low probability of becoming poor in the near future. This principle acknowledges the implications of poverty dynamics for the implementation of SAPs with a particular focus on exit conditions. This thesis characterises the poverty dynamics of beneficiary households through the estimation of a Markovian poverty transition model using data from the Familias en Accion programme. The findings from the empirical work suggest that programme participation should not end when households are non-poor, but attention must be paid to probabilities of recurrence, in order to secure non-recurrence in the near future. Taken together, the exhausted-effectiveness principle interacts with the non-recurrence of poverty principle in the sense that the first sets a maximum length of exposure to the intervention, while the second determines minimum levels of exposure.