In classic Psychological-Refractory-Period (PRP) paradigms, decreasing stimulus onset asynchronies (SOA) between the two tasks typically leads to increasing reaction time (RT2) to the second task (T2), but there is no influence on the reaction time (RT1) to the first task (T1). Traditionally, the causes of this interference have been considered to be the limitations of attentional resources or the inherent nature of central bottleneck. The PRP effect has been extensively studied and has been traditionally explained by Pashler’s response selection bottleneck (RSB) model, which proposes the processing of one task consists of three stages: (1) perceptual identification stage, which selects the task-relevant stimulus and extracts relevant attributes of said stimulus; (2) response selection (bottleneck) stage, which decides upon the appropriate motor response; (3) response execution stage, which mainly executes explicit actions. Perceptual identification stage and response execution stage were generally assumed to operate in parallel with other cognitive processing, but the bottleneck stage was assumed to operate sequentially, meaning the stage of T2 response selection was often postponed until the corresponding stage of response selection of T1 had been completed. Tombu and Jolicoeur (2003) described the Central Capacity Sharing (CCS) model, assuming that the capacity limitations of the central stags were not all or none and the processing of both tasks occurred at reducing rates due to the sharing of limited common resources. Thus, the two models had different predictions to the RT1. The RSB model assumed that T1 could get access to the bottleneck as soon as required, so RT1 remained the same at all SOA, whereas the CCS model predicted that RT1 increased with decreasing SOA. The present research used a standard PRP paradigm, in two reaction time experiments, in which participants made speeded responses to both a tone (T1) and a Stroop task (T2), the two tasks arriving in rapid succession, and stimulus intervals presenting with varying SOA. The aim is to examine the predictions of the RSB model and the CCS model, in the meantime exploring the limitations of attentional resources and the interference mechanisms in dual-task situations. The results showed that: (1) In the overlapping tasks paradigm, when T1 was processing in the central bottleneck, the response to T2 was heavily affected by the T1, and the effect of PRP was significant. Varied SOA and different difficulty T2 information substantially influenced the response selection and response execution on T1. (2) When two tasks demanded central response selection processing at the same time, the one task using more available attention resources would lead to the other task using less attention resources. The amount of resources directly determined the processing efficiency of this task. (3) There existed interference in dual-task situations, this interference not only existed in the central response selection stage, but also existed in the stage of response execution. Taken together, RT1 effects may occur when central resources are shared between the Task 1 and Task 2 processes. These results provide strong support for CCS models of dual-task interference in the overlapping tasks paradigm.