Abstract The primary charge separation in reaction centers embedded in the photosynthetic membranes induces an electric polarization in chloroplast suspensions. Photovoltages elicited by short non-saturating flashes, were observed almost 20 years ago by Witt & Zickler (1973, FEBS Lett,37, 307-310) and Fowler & Kok (1974, Progress in Photobiology, Frankfurt: Deutsche Gesellschaft). The photovoltage was interpreted as the result of the so-called "light-gradient" effect, in which the stronger excitation of the membrane facing the light source compared to the shadowed one creates a difference of dipole density in these two membranes. Owing to the anti-parallel orientation of reaction centers in opposite thylakoid membranes, a small potential difference results. It was thought that the polarity of this potential difference could be deduced from the known position of electron carriers in the photosynthetic reaction center. However, the observed polarity was often opposite to that predicted by this model. Also, the measured photovoltage amplitudes could not be quantitatively related to experimental parameters. In the present paper, we show that the "classical" explanation of the light-gradient effect does not hold true and we give an alternative explanation that is based on light propagation and interference in pigmented multilayers. A model calculation is carried out for a pair of membranes simulating stroma lamellae of chloroplasts. It predicts a wavelength-dependent light distribution as well as the polarity of the photovoltage. For low intensities, the amplitude is found to be proportional to the intensity of the incoming light, to the optical density, and to the reciprocal of the dielectric constant of the sample. When the membranes contain no chromophores or when the absorption coefficient is low, the predicted polarity is opposite to that expected from the classical picture. The model is tested with a set of experimental photovoltage data obtained at different wavelengths.