The goal of the RF program, Part B of the No Child Left Behind Act of 2001 (NCLB), is to have all children reading at grade level by the end of third grade. There is no legislative precedent for this program, focused as it is on preventing reading failure in the early elementary years (US Department of Education, April 2002). Because the design of the Reading First (RF) initiative was informed by research on effective reading instruction (National Reading Panel Report, 2000) and proposed measures to address the shortcomings of previous Title 1 and other federal initiatives, an important question is whether a program specifically focused on effective reading instruction will bring about significant improvements in students' reading achievement in the early elementary years. This study, carried out within the context of the evaluation of RF in Michigan, examines districts' criteria for selecting (and ranking) schools to receive RF funding. The authors set out not only to determine whether RDD (regression discontinuity design) is an appropriate analytic method, but also to consider other analytic approaches, given districts' selection criteria and the profiles of poverty and underachievement in RF and NonRF schools in Michigan. The analyses have documented that RDD is not a viable model to analyze the impact of RF on the achievement development in Michigan schools. District turned out to be a key variable in three respects. First of all, the probability to get funding by RF is much larger in the small school districts than in large districts. In addition, the critical level of poverty and/or low achievement varies noticeably across districts. District also differed in the way they used the poverty and achievement information in deciding which schools would participate in RF. The major finding with respect to the effects of RF is related to the different selection strategies within the district. As the analyses revealed, only those districts that based the selection on achievement criteria were able to reduce the failing rates significantly. This is remarkable, as this group includes the largest school district of Michigan, a district that in the past has stood out as having negative trends. It is likely that this school district was indirectly forced to utilize the achievement indicator to decide which school to include in RF because the poverty level is generally very high; over half of the elementary schools have a poverty rate of at least 85%. The fact that selection into RF based on low achievement turns out to be an important predictor of success is not surprising, once the nature of the intervention is taken into account. Low-achievement schools in Michigan are not automatically high poverty schools. This might suggest that many of the low-achieving schools suffer from poor implementation of the RF program components and not from detrimental conditions that are associated with poverty (e.g., low parent involvement). Although this research design does not provide a sound basis for causal inferences, these results suggest that RF might be an effective intervention for the former condition but not for the latter (see also Cortina et al., 2008). (Contains 4 tables