We study the interplay between surface roughening and phase separation during the growth of binary films. Already in 1+1 dimensions, we find a variety of different scaling behaviors, depending on how the two phenomena are coupled. In the most interesting case, related to the advection of a passive scalar in a velocity field, nontrivial scaling exponents are obtained in simulations. PACS numbers: 68.35. Rh, 05.70.Jk, 05.70.Ln, 64.60.Cn Thin solid films are grown for a variety of technological applications, using molecular beam epitaxy (MBE) or vapor deposition. In order to create materials with specific electronic, optical, or mechanical properties, often more than one type of particle is deposited. When the particle mobility in the bulk is small, surface configurations become frozen in the bulk, leading to anisotropic structures that reflect the growth history, and are different from bulk equilibrium phases [1]. Characterizing structures generated during composite film growth is not only of technological importance, but represents also an interesting and challenging problem in statistical physics.In this paper, we examine the growth of binary films through vapor deposition, and study some of the rich phenomena resulting from the interplay of phase separation and surface roughening. Simple models for layer by layer growth assume either that the probability that an incoming atom sticks to a given surface site depends on the state of the neighboring sites in the layer below [2], or that the top layer is fully thermally equilibrated [3]. Assuming that the bulk mobility is zero, once a site is occupied, its state does not change any more. If the growth rules are invariant under the exchange of the two particle types, the phase separation is in the universality class of an equilibrium Ising model. Correlations perpendicular to the growth direction are characterized by the critical exponent ν of the Ising model, and those parallel to the growth direction by the exponent νz m , with z m being the dynamical critical exponent of the Ising model. However, the layer by layer growth mode underlying these simple models is unstable, and the growing surface becomes rough. In many cases the fluctuations in the height h(x, t), at position x and time t are self-affine, with correlationswhere χ is the roughness exponent, and z h is a dynamical scaling exponent. A computer model with local sticking probabilities that allows for a rough surface was introduced in [4]. In 1+1 dimensions, the authors find phase separation into domains (with sizes consistent with the Ising model), and a very rough surface profile with sharp minima at the domain boundaries. We may ask the following questions: (1) Are the roughness exponents different at the phase transition point? (2) Are the critical exponents modified on a rough surface? We shall demonstrate that the coupling of roughening and phase separation leads to a rich phase diagram, and to nontrivial critical exponents already in 1+1 dimensions.To characterize phase separation, we introduce a...