Attributes of words, such as frequency and imageability, can influence memory for order. In serial recall, Hulme, Stuart, Brown, and Morin (Journal of Memory and Language, 49(4), 500-518, 2003) found that high-frequency words were recalled worse, and low-frequency words better, when embedded in alternating lists than pure lists. This is predicted by associative chaining, wherein each recalled list-item becomes a recall-cue for the next item. However, Hulme, Stuart, Brown, and Morin (Journal of Memory and Language, 49(4), 500-518, 2003) argued their findings supported positional-coding models, wherein items are linked to a representation of position, with no direct associations between items. They suggested their serial-position effects were due to pre-experimental semantic similarity between pairs of items, which depended on frequency, or a complex tradeoff between item- and order-coding (Morin, Poirier, Fortin, & Hulme, Psychonomic Bulletin Review, 13(4), 724-729, 2006). We replicated the smooth serial-position effects, but accounts based on pre-existing similarity or item-order tradeoffs were untenable. Alternative accounts based, on imageability, phonological and lexical neighbourhood sizes were also ruled out. The standard chaining account predicts that if accuracy is conditionalized on whether the prior item was correct, the word-frequency effect should reappear in alternating lists; however, this prediction was not borne out, challenging this retrieval-based chaining account. We describe a new account, whereby frequency influences the strengths of item-item associations, symmetrically, during study. A manipulation of word-imageability also produced a pattern consistent with item-item cueing at study, but left room for effects of imageability at the final stage of recall. These findings provide further support for the contribution of associative chaining to serial-recall behaviour and show that item-properties may influence serial-recall in multiple ways.