For portable fuel cell systems a multitude of applications have been presented over the past few years. Most of these applications were developed for indoor use, and not optimised for outdoor conditions. The key problem concerning this case is the cold start ability of the polymer electrolyte membrane fuel cell (PEMFC). This topic was first investigated by the automotive industry, which has the same requirements for alternative traction systems as for conventional combustion engines. The technical challenge is the fact that produced water freezes to ice after shut-down of the PEMFC and during start-up when the temperature is below 0 degrees C. To investigate the basic cold start behaviour isothermal, potentiostatic single cell experiments were performed and the results are presented. The cold start behaviour is evaluated using the calculated cumulated charge transfer through the membrane which directly corresponds with the amount of produced water in the PEMFC. The charge transfer curves were mathematically fitted to obtain only three parameters describing the cold start-up with the cumulated charge transfer density and the results are analysed using the statistical software Cornerstone 4.0. The results of the statistic regression analyses are used to establish a statistic-based prediction model of the cold start behaviour which describes the behaviour of the current density during the experiment. The regression shows that the initial start current mainly depends on the membrane humidity and the operation voltage. After the membrane humidity has reached its maximum, the current density drops down to zero. The current decay also depends on the constant gas flows of the reactant gases. Ionic conductivity of the membrane and charge transfer resistance were investigated by a series of ac impedance spectra during potentiostatic operation of the single cell at freezing temperatures. Cyclic voltammetry and polarisation curves between cold start experiments show degradation effects by ice formation in the porous structures which lead to significant performance loss. (c) 2005 Elsevier B.V. All rights reserved.