For asymptotically flat initial data of Einstein's equations satisfying an energy condition, we show that the Penrose inequality holds between the ADM mass and the area of an outermost apparent horizon, if the data are restricted suitably. We prove this by generalizing Geroch's proof of monotonicity of the Hawking mass under a smooth inverse mean curvature flow, for data with non-negative Ricci scalar. Unlike Geroch we need not confine ourselves to minimal surfaces as horizons. Modulo smoothness issues we also show that our restrictions on the data can locally be fulfilled by a suitable choice of the initial surface in a given spacetime. An important issue in General Relativity is the "cosmic censorship conjecture" which reads, roughly speaking, that singularities (which necessarily develop in gravitational collapse in the future of apparent horizons) are always separated from the outside world by event horizons. Penrose gave a heuristic argument which showed that, for collapsing shells of particles with zero rest mass, cosmic censorship would imply the inequalitybetween the total mass (the ADM-mass) M of the spacetime and the area A of an apparent horizon H [1]. While for such shells inequality (1) was recently proven by Gibbons [2], there has remained the challenge of proving (independently of cosmic censorship) the Penrose inequality (PI) (1) in the "general case" characterized below.We consider a spacetime (M, 4 g µν ) and a corresponding initial data set (N , g ij , k ij ), i.e. a smooth 3-manifold N (which can be embedded in M), a positive definite metric g ij and a symmetric tensor k ij (the second fundamental form of the embedding). To be compatible with Einstein's equations on M, these quantities satisfy the constraintswhere D i is the covariant derivative and R the Ricci scalar on N , k = g ij k ij , and ρ and j i are the energy density and the matter current, respectively. We take the data to be asymptotically flat and to satisfy the dominant energy condition (which implies that ρ ≥ |j|). Furthermore, we assume that the boundary of N (if non-empty) is a "future apparent horizon" H (called a "horizon" from now on) which is a 2-surface defined by the property that all outgoing future directed null geodesics (in M) orthogonal to H have vanishing divergence, i.e. θ + = 0, and that the divergence of the outgoing past null geodesics is non-negative, i.e. θ − ≥ 0. (Past apparent horizons are defined similarly, and satisfy analogous theorems). If θ + = θ − = 0 on H, the horizon is an extremal surface of (N , g ij ) and the outermost extremal surface in an asymptotically flat space must be minimal.As an idea for proving the positive mass theorem (PMT) Geroch considered an asymptotically flat Riemannian manifold (N , g ij ) with non-negative Ricci scalar R (which naturally arises from initial data sets described above by restricting k ij suitably) and assumed that on (N , g ij ) there is a smooth "inverse mean curvature flow" (IMCF). This means that one can write the metric g ij as(where A, B = 2, 3), with smoo...