We explored the impact of operand format (digit, word, pseudo-homophone) on single-digit addition and multiplication. Format manipulations are of theoretical interest because models of arithmetic knowledge differ with respect to predicted format effects. Latencies were shortest when operands were digits, and longest when they were pseudo-homophones. However, there was also an interaction of problem size with format: problem size effects were smaller in the pseudo-homophone condition relative to the digit condition, and were larger in the word format condition relative to the digit condition. We discuss our results with respect to the following question: Can this interaction be attributed (solely) to an encoding phase of processing, or might it (also) arise from a solution phase?