Two experiments examined transfer of mathematical problem-solving procedures learned in content-rich quantitative domains (e.g., physics, finance) to isomorphic algebra word problems dealing with other contents. In spite of content-specific embedding, many high school and college students exhibited spontaneous transfer when the variables in the learned and in the transfer problems represented similar types of quantities (e.g., speed and typing rate). Spontaneous transfer to structurally isomorphic problems with variables representing different types of quantities (e.g., speed and salary) was blocked. Protocol analyses showed that for matching quantities, transfer solutions were straightforward applications of the learned methods, whereas transfer to problems with nonmatching quantities demanded a hint for retrieval followed by an effortful process of abstraction and analogical mapping. The results suggest a complex interrelation between content and structure: Many features of the embedding content may be "screened out" as irrelevant, but content features used for interpretation of variables (e.g., intensive vs. extensive) may affect both access and use.