The spatial distribution of the actinic flux is investigated for a realistic scattering and absorbing atmosphere with embedded two-dimensional (2-D) clouds. Three different models are intercompared for computing the actinic flux: MCC4, a versatile Monte Carlo code, SHDOM, a freely available code for solving the multidimensional integral form of the radiative transfer equation, and DISORT, a widely used discrete ordinate code for plane-parallel media. Three different cloud scenarios are chosen; plane-parallel clouds, internally homogeneous rectangular cloud bands, and a realistically variable stratocumulus cloud field. For scattered cloud fields the results show that both Rayleigh scattering as well as ground reflection significantly smooth out the actinic flux field. For cloud fields with realistic spatially inhomogeneous liquid water distributions, local maxima of the actinic flux are strongly correlated with the corresponding maxima in cloud water. The spatial patterns of the actinic flux field as computed with MCC4 and SHDOM agree very well. The horizontally averaged flux results of MCC4 and SHDOM are within 1--2% of each other. For the 2-D cloud cases considered, SHDOM turned out to be 1 to 2 orders of magnitude faster than MCC4. For the stratocumulus cloud photodissociation coefficients for NO2 and O(1D) were computed with SHDOM and compared with the plane-parallel (PP), independent pixel (IP), and locally smoothed IP (SIP) approximations. The PP is characterized by a general underestimation of the photodissociation coefficients in and below the cloud, whereas the IP reproduces horizontal averages satisfactorily well. Local biases of the IP near cloud top can be significantly reduced by employing the SIP. ¿ 1999 American Geophysical Union