Much CA research is grounded in specimen collections, which are numerically modest by the standards of survey research or corpus linguistics, but substantial relative to observational fieldwork. The appeal of collection-based methods is that they afford some of the advantages of context-sensitive case analysis, while also enabling the development of accounts whose generality may be tested across a number of cases. They have a particular utility for the investigation of novel phenomena in areas whose elementary units and basic organizational forms are not well-understood. This chapter reflects on key issues involved in both assembling and working through specimen collections. Regarding the assembly of cases, it is argued that researchers should cast a wide net across a diversity of data sources, taking care to avoid allowing hunches or hypotheses to gain a controlling influence over data collection. Regarding the investigation of patterns across cases, the discussion touches on the utility of single case analyses, systematic reviews of the entire collection, and various approaches to dealing with anomalous cases. The chapter concludes with a discussion of the limitations of prototypical specimen collections, identifying conditions when it may be advisable to augment a collection by adding cases beyond the target phenomenon.