Poverty is a major cause of a number of social problems, political, economic and particularly in developing countries. Poverty reduction is the key factor that must be done to achieve the targets in the Millennium Development Goals (MDGs) by 2015 government assistance program for the reduction of poverty that exist today have not been able to reduce the number of poor households in Jombang, whereas Jombang have a positive economic growth rate from year to year. Thus, in order to support effective and efficient government then carried out a classification based on the household assistance package that is expected by the poor households in Jombang. Poor households as the object of observation is divided into two according to their poverty status, namely Poor Households (RTM) and Chronically Poor Households (RTSM). Classification is done by using the CART approach is then combined with the method of the Random Forest to obtain a more accurate and stable classification. Classification results using Random Forest CART stated that the expected assistance package by RTSM determined by income per month, place of dispocing of fecal water, type of floor building, and source of drink water. While, the most important characteristic of RTM classification income per month, place of dispocing of fecal water, source of drink water, and large of house building. The classification result from Random Forest CART method produces a total accuracy rate of 99.50 percent for RTSM and 98.33 percent for RTM. The mapping of households in Jombang stated that cash money, cattle, and modal were the most expexted assistance package.