In this paper we focus on {\em serial episodes}, that are sequences of event types extracted from single or multiple input sequences, and that reflect a qualitative relation {\em is-followed-by} between the event types. Episodes have natural applications into several domains, including for instance the analysis of business time series, medical data, geophysical data and also alarm log analysis for network monitoring (especially in telecommunications). However, in many applications episodes clearly show some limitations, due to the fact that the information provided by the {\em is-followed-by} relation is not always enough to properly characterize the phenomena at hand. This, in particular, pulls our research toward the refinement of episodes to incorporate quantitative temporal information, able to describe the time intervals observed for the {\em is-followed-by} relation. We proposed a refinement of episodes called {\em quantitative episodes}, that provides quantitative temporal information in a readable, tree-based graphically representable form. These quantitative episodes describe the main groups of homogeneous behaviors within the occurrences of each episode, according to the elapsed times between the consecutive event types of the episode. Moreover, they are not provided in an isolated way, but in trees giving a global view of how the occurrences of the corresponding episode differentiate in homogeneous groups along the elements of the pattern. From a computational point of view, the main interest of the quantitative episodes is that they can be mined in a sound and complete way without increasing the cost of extractions significantly when compared to extractions of episodes alone. This is achieved through an extraction algorithm that tightly integrates episode extraction with a computationally reasonable analysis of temporal quantitative information.