We present an experimental study of a buoyancy-driven, low-Reynolds-number (Re < 1) exchange flow of two Newtonian fluids in a vertical cylindrical pipe (length 1 m and diameter 38.4 mm) connecting two fluid reservoirs. The denser, more viscous fluid was golden syrup and the less dense, less viscous fluid was a golden syrup–water solution; the ratio of the viscosities of the two fluids (β) ranged from 2 to 1180. Flows were initiated by removing a bung in the base of the upper reservoir or sliding out a gate positioned at the top, middle or bottom of the pipe. We observe the flows over long time durations (up to 356 h), and define the development of the flow with reference to a non-dimensional time (τ). The initial transient development of the flow was dependent on which of the two fluids initially filled the pipe, but this did not systematically affect the flow regime observed at τ ≫ 1. Two distinct flow regimes were observed: axisymmetric core-annular flow (CAF), in which the less viscous fluid occupies a cylindrical core and the denser fluid flows downwards in an annulus, and side-by-side (SBS) flow where both fluids are in contact with the pipe and there is a single interface between them. CAF formed at β ≥ 75 and SBS flow at β ≤ 117. In several experiments, for 5 ≤ β ≤ 59, a slowly developing transitional SBS (TSBS) flow was observed where SBS flow and CAF occurred simultaneously with SBS in the lower portion of the pipe; SBS existed throughout most of the pipe and in one case grew with time to entirely fill the pipe. Velocity profiles determined by tracking tracer particles show that the observed CAFs are adequately described by the formulation of Huppert & Hallworth (J. Fluid Mech., vol. 578, 2007, pp. 95–112). Experimental SBS velocity profiles are not well produced by the formulation of Kerswell (J. Fluid Mech., 10.1017/jfm.2011.190), possibly because the latter is restricted to flows whose cross-section has an interface of constant curvature. Despite the variations in flow regime, volume fluxes can be described by a power-law function of β, Q1 = 0.059 β−0.74. A comparison of experimental data with the theoretical approaches of Huppert Hallworth (2007) and Kerswell (2011) indicates that fluids are not arranged in the regime that maximises volume flux (e.g. SBS or CAF), nor do they adopt the geometry that maximises volume flux within that particular regime.