This study assessed the usefulness of Multivariate Adaptive Regression Splines (MARS) in identifying household poverty correlates in the Philippines. Using data from the 2003 Family Income and Expenditure Survey (FIES) and 2005 Community-Based Monitoring System (CBMS) for Pasay City, MARS was compared with Logistic Regression (LR) based on various classification accuracy measures. Based on the results, models produced by MARS are more parsimonious yet contains theoretically and empirically sound set of household poverty correlates and have relatively higher accuracy in terms of predicting the poor or the potential program beneficiaries, particularly under the following conditions: (i) large variations among the independent variables, leading to good spread of groups of observations; (ii) low proportion of missing observations; (iii) large number of observations; (iv) lower interaction order (when the number of candidate predictors is relatively small), and; (v) lower penalty values (also when the number of candidate predictors is relatively small). The findings suggest that MARS can be a better alternative to LR for a more efficient and effective implementation of a proxy means test, especially in developing countries like the Philippines where budget for poverty alleviation programs is limited.