The Isle au Haut Igneous Complex provides a unique opportunity to examine in detail the in situ physical and chemical interactions between contemporaneously emplaced mafic and silicic magmas. The complex contains a 600 m thick sequence of 11 alternating layers of gabbro and diorite (typically 15–40 m thick). Purely on the basis of density contrasts (2·65 g cm−3 gabbro vs 2·55 g cm−3 diorite), the entire system should have undergone wholesale instability and mixing; it is instead arrested in a grossly unstable state of interaction while molten. Chilled margins along the lower contacts of the gabbros and structural integrity of the diorite layers indicate that near-liquidus gabbroic magma invaded partly crystalline, cooler diorite. Mineral assemblages, chemical analyses, and phase equilibria calculations indicate initial temperatures during emplacement of ∼1180°C (gabbro) and ∼1000°C (diorite). Conductive thermal models yield solidification timescales of 15–60 years for single gabbro layers and about a thousand years for the entire complex. There is ample evidence for two phases of small-scale interfacial Rayleigh–Taylor type instabilities of dioritic melt into the gabbros. Phase I occurred immediately upon gabbro emplacement whereas evenly spaced, slender more silicic pipes represent a much later stage (Phase II). Pipe geometry and spacing, estimated viscosities of the gabbroic magma and silicic melt, and the sudden increase in silica near the upper contact of the diorite, all indicate a thin (∼18–53 cm) buoyant layer at the upper contact of the diorite as the source of the pipes. Compaction of the diorite produced this layer over a period of about 10 years. Simultaneous solidification along the lower contact of the overlying gabbro, thickening inwards, increased viscosity enough to arrest pipe ascent after a few meters. Crystal size distribution analyses of the gabbro layers yield crystal growth rates [Go = (2 − 4) × 10−10 cm s−1] and nucleation rates (Jo = 10−5–10−6 cm−3 s−1) indicative of conductive cooling coupled with some sluggish convective stirring owing to collapse of the roof-ward gabbro solidification fronts. Were the complex larger, with a much longer solidification time, all this evidence would have been lost, thus suggesting that in larger systems similar processes may commonly take place leaving little direct evidence of their operation apart from the ultimate final petrological product.