The problem-size effect in simple additions, that is the increase in response times (RTs) and error rates with the size of the operands, is one of the most robust effects in cognitive arithmetic. Current accounts focus on factors that could affect speed of retrieval of the answers from long-term memory such as the occurrence of interference in a memory network or the strength of memory traces that would differ from problem to problem. The present study analyses chronometric data from a sample of 91 adults solving very small additions (operands from 1 to 4) that are generally considered as being solved by retrieval. The results reveal a monotonic linear increase in RTs with the magnitude of both operands. This pattern is at odds with the retrieval-based accounts of the problem-size effect and challenges the well-established view that small additions are solved through retrieval of the answer from long-term memory. Our results are more compatible with the hypothesis that even very small additions are solved using compacted fast procedures that scroll an ordered representation such as a number line or a verbal number sequence. This interpretation is corroborated by the analysis of individual differences.