Production performance of grass-fed dairy cows is often lower than expected from the estimated energy supply. To explain the overestimation of the energy content of grass for dairy cows, data from energy balance trials from three different laboratories (Wageningen, Lelystad and Hillsborough) were collected. The trials in Wageningen and Lelystad were carried out in the 1970s and those in Hillsborough in the 1990s. Regression analyses were carried out with the complete data set as well as per laboratory to identify differences per laboratory. Average net maintenance requirements per kg3/4 (NEm) were 0.573 MJ, whereas the efficiency of metabolizable energy utilization for lactation (k1) was 0.777. When NEm was fixed at the presently used value of 0.293 MJ/kg3/4, k1 was 0.60. Between-laboratory NEm varied between 0.294 (Lelystad) and 0.786 MJ/kg3/4 (Hillsborough), whereas k1 varied between 0.57 (Lelystad) and 0.84 (Hillsborough). For Wageningen and Hillsborough, NEm was high, whilst k1 was also high. With the intercept fixed at 0.293 MJ/kg3/4, efficiency varied between 0.53 (Hillsborough) and 0.62 (Wageningen). The k1 and NEm are interrelated. Based on these data we surmise that the maintenance requirements for grass-fed dairy cows are 10% higher than presently assumed, with no change in k1. © 2002 Elsevier Science B.V. All rights reserved.