The Australia Telescope was used in March–April 2005 to observe the 1.384 and 2.368-GHz emissions from the RS CVn binary HR 1099 in two sessions, each of 9-h duration and 11 days apart. Two intervals of highly polarised emission, each lasting 2–3 h, were recorded. During this coherent emission we employed a recently installed facility to sample the data at 78-ms intervals to measure the fine temporal structure and, in addition, all the data were used to search for fine spectral structure. We present the following observational results: (1) ~100% left-hand circularly polarised emission was seen at both 1.384 and 2.368 GHz during separate epochs; (2) the intervals of highly polarised emission lasted for 2–3 h on each occasion; (3) three 22-min integrations made at 78-ms time resolution showed that the modulation index of the Stokes V parameter increased monotonically as the integration time was decreased and was still increasing at our resolution limit; (4) the extremely fine temporal structure strongly indicates that the highly polarised emission is due to an electron-cyclotron maser operating in the corona of one of the binary components; (5) the first episode of what we believe is ECME (electron-cyclotron maser emission) at 1.384 GHz contained a regular frequency structure of bursts with FWHM ~48 MHz, which drifted across the spectrum at ~0.7 MHz min^–1. Our second episode of ECME at 2.368 GHz contained wider-band frequency structure, which did not permit us to estimate an accurate bandwidth or direction of drift; (6) the two ECME events reported in this paper agree with six others reported in the literature in occurring in the binary orbital phase range 0.5–0.7; (7) in one event of 8-h duration, two independent maser sources were operating simultaneously at 1.384 and 2.368 GHz.

We discuss two kinds of maser sources that may be responsible for driving the observed events that we believe are powered by ECME. One is based on the widely reported ‘loss-cone anisotropy’, the second on an auroral analogue, which is driven by an unstable ‘horseshoe’ distribution of fast-electron velocities with respect to the magnetic field direction. Generally, we favour the latter, because of its higher growth rate and the possibility of the escape of radiation which has been emitted at the fundamental electron cyclotron frequency. If the auroral analogue is operating, the magnetic field in the source cavity is ~500 G at 1.384 GHz and ~850 G at 2.368 GHz; the source brightness temperatures are of the order T_B ~ 10¹⁵ K.

We suggest that the ECME source may be an aurora-like phenomenon due to the transfer of plasma from the K2 subgiant to the G5 dwarf in a strong stellar wind, an idea that is based on VLBA maps showing the establishment of an 8.4 GHz source near the G5 dwarf at times of enhanced radio activity in HR 1099.