Traditional models of human memory have postulated the need for a brief phonological or verbatim representation of verbal input as a necessary gateway to a higher level conceptual representation of the input. Potter has argued that meaningful sentences may be encoded directly in a conceptual short-term memory (CSTM) running parallel in time to such a phonological store. The primary aim of the current study was to evaluate two main tenets of the CSTM model: that linguistic context biases selection of information entering the conceptual store, and that information not integrated into a coherent structure is rapidly forgotten. Results confirmed these predictions for spoken sentences heard by both young and older adults, supporting the generality of the model and suggesting that CSTM remains stable in normal aging.