Perovskite materials have generated significant interest from academia and industry as a potential component in next-generation, high-efficiency, low-cost, photovoltaic (PV) devices. The record efficiency reported for perovskite solar cells has risen rapidly, and is now more than 22%. However, due to their complex dynamic behaviour, the process of measuring the efficiency of perovskite solar cells appears to be much more complicated than for other technologies. It has long been acknowledged that this is likely to greatly reduce the reliability of reported efficiency measurements, but the quantitative extent to which this occurs has not been determined. To investigate this, we conduct the first major inter-comparison of this PV technology. The participants included two labs accredited for PV performance measurement (CSIRO and NREL) and eight PV research laboratories. We find that the inter-laboratory measurement variability can be almost ten times larger for a slowly responding perovskite cell than for a control silicon cell. We show that for such a cell, the choice of measurement method, far more so than measurement hardware, is the single-greatest cause for this undesirably large variability. We provide recommendations for identifying the most appropriate method for a given cell, depending on its stabilisation and degradation behaviour. The results of this study suggest that identifying a consensus technique for accurate and meaningful efficiency measurements of perovskite solar cells will lead to an immediate improvement in reliability. This, in turn, should assist device researchers to correctly evaluate promising new materials and fabrication methods, and further boost the development of this technology.