Assuming an exponential model, fire growth rates are estimated in this paper for four types of occupancies - railway properties, public car parks, road tunnels and power stations. The results are based on data for area damaged and duration of burning in actual fires contained in fire statistics. It is shown clearly that the maximum value (upper confidence limit) of growth rates in individual fires is distinct from the maximum value of the average growth rate in all fires. The former maximum is considerably higher than the latter and represents more realistically the worst case scenario. The statistically-determined fire growth rate discussed in this paper does not relate to any particular object. It is a combined (average) estimate for several objects involved in a fire and takes into account the stochastic (random) nature of fire spread from object to object. Such an estimate allowing for uncertainties governing the development of a real fire can not be provided, at present, by any deterministic model based on heat output data from experiments.