Sequences of residence times (RTs) associated with the escape from metastable states are observed in many fields. Here we study analytically and numerically the correlations among RTs for a bistable stochastic system driven by dichotomous noise. Our theory predicts an oscillatory behavior of the correlations with respect to the lag between RTs. Correlations vanish at all lags if the switching rate matches the hopping rate of the unperturbed system. It is also shown that RT correlations may reveal features of the driving which are not present in the single-RT statistics. Not much attention has been paid to the fact that RTs are most naturally observed in sequences: the subsequent epochs spent by a Brownian particle in the left and right wells of a quartic potential, the time spent in the neighboring minima of a periodic potential, or the periods spent near the resting state of an excitable system [7]-all of these residence times are commonly measured (in experiments or in a model simulation) as an ordered sequence fI k g, with k 0; 1; . . . . These RTs will be only in the most simple cases mutually independent. In any case involving external driving, inertia, delayed feedback, etc., we expect that RT correlations will emerge. Except for certain neuron models where sequences of interspike intervals correspond to sequences of first-passage times [8][9][10], however, such correlations in the RT sequence have not been studied to the best of our knowledge.In this Letter, we study RT correlations in a simple setup that is analytically tractable. For a bistable system driven by a thermal and by a dichotomous noise, the standard statistics of the RT (i.e., the RT density) will be contrasted with the surprising features of the RT correlation statistics. We show that, besides the expected behavior of correlations at the small switching rate of the dichotomous driving, we find nonintuitive effects such as the vanishing of linear correlations when the driving's switching rate matches the unperturbed escape rate of the system and an inversion of the RT correlation's sign for an even larger switching rate of the driving. We also demonstrate how input parameters can be inferred from the RT correlations if the driving is slow and weak.We consider the dynamics of an overdamped bistable system driven by a dichotomous (telegraph) noise: