A generalization of the microcanonical ensemble suggests a simple strategy for the simulation of first order phase transitions. At variance with flat-histogram methods, there is no iterative parameters optimization, nor long waits for tunneling between the ordered and the disordered phases. We test the method in the standard benchmark: the Q-states Potts model (Q = 10 in 2 dimensions and Q = 4 in 3 dimensions), where we develop a cluster algorithm. We obtain accurate results for systems with 10 6 spins, outperforming flat-histogram methods that handle up to 10 4 spins. No cure is known for ECSD in canonical simulations (cluster methods [3, 4] do not help), which motivated the invention of the multicanonical ensemble [5]. The multicanonical probability for the energy density is constant, at least in the energy gap e o < e < e d (e o and e d : energy densities of the coexisting low-temperature ordered phase and high-temperature disordered phase), hence the name flat-histogram methods [5,6,7,8]. The canonical probability minimum in the energy gap (∝ exp[−ΣL D−1 ]) is filled by means of an iterative parameter optimization.In flat-histogram methods the system performs an energy random walk in the energy gap. The elementary step being of order L −D (a single spin-flip), one naively expects a tunneling time from e o to e d of order L 2D spinflips. But the (one-dimensional) energy random walk is not Markovian, and these methods suffer ECSD [10]. In fact, for the standard benchmark (the Q = 10 Potts model [9] in D = 2), the barrier of 10 4 spins was reached in 1992 [5], while the largest simulated system (to our knowledge) had 4 × 10 4 spins [6].ECSD in flat histogram simulations is probably understood [10]: on its way from e d to e o , the system undergoes several (four in D = 2) "transitions". First comes the condensation transition [10,11], at a distance of order L −D/(D+1) from e d , where a macroscopic droplet of the ordered phase is nucleated. Decreasing e, the droplet grows to the point that, for periodic boundary conditions, it reduces its surface energy by becoming a strip [12], see Fig. 2 (in D = 3, the droplet becomes a cylinder, then a slab [13]). At lower e the strip becomes a droplet of disordered phase. Finally, at the condensation transition close to e o , we encounter the homogeneous ordered phase.Here we present a method to simulate first order transitions without iterative parameter optimization nor energy random walk. We extend the configuration space as in Hybrid Monte Carlo [14]: to our N variables, σ i (named spins here, but they could be atomic positions) we add N real momenta, p i . The microcanonical ensemble for the {σ i , p i } offers two advantages. First, microcanonical simulations [15] are feasible at any value of e within the gap. Second, we obtain FluctuationDissipation Eqs. (5-8) where the (inverse) temperaturê β, a function of e and the spins, plays a role dual to that of e in the canonical ensemble. The e dependence of the mean value β e , interpolated from a grid as it is almost cons...