The divergence criterion allows detection in advance runaway initiation. When divergence (div) is greater than zero, this indicates that the reactor is moving to a dangerous state. In this work, a series of experiments have been carried out in a pilot-plant reactor, using the esterification reaction between 2-propanol and propionic anhydride, in order to assess the performances of a developed algorithm to calculate div online using temperature or pressure data. Furthermore, the model developed for this process allows a comparison between simulated and experimentally reconstructed divergences. The results show that a good agreement is obtained for the runaway experiments, whereas for near-runaway conditions, the presence of "noise" in the data shows the necessity to establish a limit value for the divergence, i.e., div > lim. In any case, we show that this methodology can be applied online to detect anomalous behavior in chemical reactors.