Does program performance vary with program type? It has long been argued that different types of public programs face different management challenges, and that these differences may affect program performance. To study this question, we use data from OMB’s Program Assessment Rating Tool (PART). Several characteristics of the PART dataset are useful for understanding the relationship between program type and program performance. First, programs were classified into seven types under PART (direct federal, credit, research and development, competitive grant, capital assets and service acquisition, regulatory, and block/formula grant). Second, PART generated performance scores under four categories (program purpose and design, strategic planning, program management, and program results).  Thus, it is straightforward to generate data from PART on program type and program performance. But modeling the relationship between program types and performance requires additional variables. We focus specifically on goal ambiguity variables, on the assumption that some types of programs face greater goal ambiguity, and that goal ambiguity in part explains PART performance scores. We use three measures of goal ambiguity recoded from the PART data. Target-specification ambiguity is measured as the proportion of a program’s stated goals that do not have specified performance targets.  Evaluation ambiguity is measured as the proportion of performance measures that are outputs rather than outcomes. Timeline ambiguity is measured as the proportion of performance measures with clear timelines for meeting targets. We hypothesize that programs vary with regard to each type of ambiguity, and that each type of ambiguity in turn affects program performance. We use structural equation modeling to estimate the relationships between program type, the three measures of goal ambiguity, and PART performance scores. The preliminary results show statistically significant relations between program type and performance. Some dimensions of goal ambiguity affect performance scores for some types of programs but not for others. In addition to presenting the empirical findings, the paper will discuss theoretical implications and directions for future research.