IntroductionBarbourfl] invented an ingenious method of establishing the asymptotic distribution of the number X of specified subgraphs of a random graph. The novelty of his method relies on using the first two moments of X only, despite the traditional method of moments that involves all moments of X (compare [8,10,11,14]). He also adjusted that new method for counting isolated trees of a given size in a random graph. (For further applications of Barbour's method see [4] and [10].) The main goal of this paper is to show how this method can be extended to a general setting that enables us to derive asymptotic distributions of subsets of vertices of a random graph with various properties.Throughout the paper we focus our attention on those properties that are determined by the presence or absence of the edges having at least one endpoint in the subset, i.e. lying totally inside it or belonging to its edge-cut. We call such properties semi-induced. For instance, the properties of being an independent set, dominating set, or the vertex set of a connected component are of this type. Also, being a vertex of a given degree is a semi-induced property.In the next section we formally define various classes of properties of subsets of vertices of a graph. Then we develop and refine Barbour's original argument and obtain several estimates for the total variation distance between the distributions of random variables in question and appropriate Poisson distributions.Section 3 is devoted to applications of those general results to a random graph K n v , obtained by independent deletion of the edges of a complete graph on n vertices with probability 1 -p. In particular, we state conditions under which the number of vertices of K n with a given degree has asymptotically normal distribution. Our result is a substantial extension of the normal phase that has been previously found (see, for example, [13]).Finally, using our estimates from Section 2 we establish the asymptotic distributions of the number of independent, 2-independent, and dominating sets of a given size in K n p . Some of these results extend and strengthen earlier results of Burtin (see [5,6]).We write X n^->Po(A) and X n^~> N(0,1) when a sequence of random variables {X n } converges in distribution to the Poisson distribution with expectation A or to the standard normal distribution, respectively. Given a random variable X, EX stands for its expectation, var X for its variance, and X for its standardization, i.e.