This experimental study examines relationships between alternative evolution paths of basaltic liquids (the so-called Bowen and Fenner trends), and silicate liquid immiscibility. Synthetic analogues of natural immiscible systems exhibited in volcanic glasses and melt inclusions were used as starting mixtures. Conventional quench experiments in 1 atm gas mixing furnaces proved unable to reproduce unmixing of ferrobasaltic melts, yielding instead either turbid, opalescent glasses, or crystallization of tridymite and pyroxenes. In contrast, experiments involving in situ high-temperature centrifugation at 1000g (g = 9·8 m/s2) did yield macroscopic unmixing and phase separation. Centrifugation for 3–4 h was insufficient to complete phase segregation, and resulted in sub-micron immiscible emulsions in quenched glasses. For a model liquid composition of the Middle Zone of the Skaergaard intrusion at super-liquidus temperatures of 1110–1120°C, centrifugation produced a thin, silicic layer (64·5 wt% SiO2 and 7·4 wt% FeO) at the top of the main Fe-rich glass (46 wt% SiO2 and 21 wt% FeO). The divergent compositions at the top and bottom were shown in a series of static runs to crystallize very similar crystal assemblages of plagioclase, pyroxene, olivine, and Fe–Ti oxides. We infer from these results that unmixing of complex aluminosilicate liquids may be seriously kinetically hampered (presumably by a nucleation barrier), and thus conventional static experiments may not correctly reproduce it. In the light of our centrifuge experiments, immiscibility in the Skaergaard intrusion could have started already at the transition from the Lower to the Middle Zone. Thus, magma unmixing might be an important factor in the development of the Fe-enrichment trend documented in the cumulates of the Skaergaard Layered Series.