This article presents a new model, the time-based resource-sharing model, that explains working memory spans in adults. The model suggests that both processing and maintenance components of working memory tasks require attention, and memory traces decay when attention is diverted. The cognitive load of the processing component is determined by the number of memory retrievals and the time allowed for them. Seven experiments showed that working memory spans vary with this cognitive load. Working memory is generally viewed as a system coordinating processing and storage. Adults' performance on reasoning tasks is impaired by concurrent memory load, especially with more difficult tasks. Baddeley and Hitch (1974) proposed a central executive system responsible for the processing-storage tradeoff. Resource-sharing models led to complex span tasks that involve both processing and storage. These tasks have been highly predictive of performance in reading comprehension, reasoning, and complex learning. However, Towse and Hitch (1995) argued that working memory spans depend on the duration of the processing component, not difficulty. Their task-switching model suggested that spans rely on retention periods, not cognitive demand. Barrouillet and Camos (2001) challenged this by showing that cognitive cost also affects spans. They proposed a time-based resource-sharing model that integrates time constraints and resource-sharing processes into working memory tasks, providing a metric of cognitive load (CL). The model was tested in experiments using new tasks that controlled for task nature and duration. In the first series, adults' spans mainly depended on the cognitive cost of the processing component, while total duration had a moderate effect. In the second series, evidence showed that working memory span depends on both the number of memory retrievals and the time allowed for them. The time-based resource-sharing model assumes that both processing and maintenance require attention, and memory traces decay when attention is switched away. Retrievals are constrained by a central bottleneck, affecting maintenance. The model predicts that processing components involving retrievals are more disruptive for memory item maintenance. The model was tested in experiments comparing operation span, reading operation span, and baba span tasks. The continuous operation span task involved solving simple operations, while the reading operation span task involved reading answers. The baba span task involved repeating a syllable. The results showed that the continuous operation span was lower than the reading operation span, which was lower than the baba span, supporting the resource-sharing hypothesis. The model also showed that cognitive load depends on the number of retrievals and time allowed, with longer durations leading to lower spans. The model predicts that tasks with longer centrally demanding periods result in lower recall performance. The results support the resource-sharing hypothesis and show that working memory spans depend on the nature of the processing component. The model provides a metric of cognitive load and explains the effects of time and resource constraints on working memory spans.This article presents a new model, the time-based resource-sharing model, that explains working memory spans in adults. The model suggests that both processing and maintenance components of working memory tasks require attention, and memory traces decay when attention is diverted. The cognitive load of the processing component is determined by the number of memory retrievals and the time allowed for them. Seven experiments showed that working memory spans vary with this cognitive load. Working memory is generally viewed as a system coordinating processing and storage. Adults' performance on reasoning tasks is impaired by concurrent memory load, especially with more difficult tasks. Baddeley and Hitch (1974) proposed a central executive system responsible for the processing-storage tradeoff. Resource-sharing models led to complex span tasks that involve both processing and storage. These tasks have been highly predictive of performance in reading comprehension, reasoning, and complex learning. However, Towse and Hitch (1995) argued that working memory spans depend on the duration of the processing component, not difficulty. Their task-switching model suggested that spans rely on retention periods, not cognitive demand. Barrouillet and Camos (2001) challenged this by showing that cognitive cost also affects spans. They proposed a time-based resource-sharing model that integrates time constraints and resource-sharing processes into working memory tasks, providing a metric of cognitive load (CL). The model was tested in experiments using new tasks that controlled for task nature and duration. In the first series, adults' spans mainly depended on the cognitive cost of the processing component, while total duration had a moderate effect. In the second series, evidence showed that working memory span depends on both the number of memory retrievals and the time allowed for them. The time-based resource-sharing model assumes that both processing and maintenance require attention, and memory traces decay when attention is switched away. Retrievals are constrained by a central bottleneck, affecting maintenance. The model predicts that processing components involving retrievals are more disruptive for memory item maintenance. The model was tested in experiments comparing operation span, reading operation span, and baba span tasks. The continuous operation span task involved solving simple operations, while the reading operation span task involved reading answers. The baba span task involved repeating a syllable. The results showed that the continuous operation span was lower than the reading operation span, which was lower than the baba span, supporting the resource-sharing hypothesis. The model also showed that cognitive load depends on the number of retrievals and time allowed, with longer durations leading to lower spans. The model predicts that tasks with longer centrally demanding periods result in lower recall performance. The results support the resource-sharing hypothesis and show that working memory spans depend on the nature of the processing component. The model provides a metric of cognitive load and explains the effects of time and resource constraints on working memory spans.