In formulating social policy the administrative arm of government relies heavily on number-based significations of knowledge, such as needs indicators and performance measures. Relying on numbers increases administrators' confidence in their decisions and shifts responsibility for error away from the decision-maker and towards the numbers. A close examination of the technology of social quantification reveals instability in many of the definitions and codes that needs analysts and program evaluators adopt when numerically inscribing social entities. To deal with these risks, bureaucracies must establish ways of explicitly assessing the uncertainty, imprecision and social construction that often lies behind the evidence presented as numbers, evidence that can easily be accepted on face value and be turned uncritically into decision-making rationales.