Abstract: In this paper the limiting properties of relative entropy densities （（f_n）_n≥1 are discussed. Whence, at sufficient condition that （X_n）_n>1 follows the strong law of large numbers is derived. Definition. A binary information source is a sequence of random variables X_n,n≥1, such that 1. Each X_n takes values in 0, 1; 2. For all ,n≥1 p（x₁, …, x_n）=PX₁,…, X_n=x_n＞0, x_i∈0, 1, 1≤i≤n In the following theorems we assume that X_n,n≥1 is an arbitrary binary source S_n（ω） =X₁（ω）+…+X_n（ω）,H（x, 1-x） is the entropy of Bernoulli distribution, that is, H（x, 1-x）= -x log x-（1-x） log （1-x） where the logarithm base is 2, and \varphi_n(\omega)=\frac1n \sum_i=1^n\leftX_i \log p_i+\left(1-X_i\right) \log \left(1-p_k\right)\right-\frac1n \log p\left(X_1, \cdots, X_n\right) where p_n∈（0, 1）, n=1, 2, … is a given sequence of real numbers. Theorem 1. \limsup _n \rightarrow \infty \varphi_n(\omega)<0 \quad a.e. Corollary 1. \undersetn \rightarrow \infty\lim \sup \left-\frac1n \log p\left(X_1, \cdots, X_n\right)\right<1 a.e. Corollary 2. If \undersetn \rightarrow \infty\lim \sup \fracS_n(\omega)n \geqslant p \quad \text a.e., \quad \frac12 \leqslant p \leqslant 1 or \limsup _n \rightarrow \infty \fracS_n(\omega)n \leqslant p \quad \text a.e., 0 \leqslant p \leqslant \frac12 then \limsup _n \rightarrow \infty\left-\frac1n \log p\left(X_1, \cdots, X_n\right)\right \leqslant H(p, 1-p) \text a.e. Theorem 2. If \liminf _n \rightarrow \infty\left-\frac1n \log p\left(X_1, \cdots, X_n\right)\right \geqslant C \quad \text a.e. then \liminf _n \rightarrow \infty H\left(\fracS_n(\omega)n, \quad 1-\fracS_n(\omega)n\right) \geqslant C \quad \text a.e. Theorem 3. Letting \begingathered D=\left\\omega: \undersetn \rightarrow \infty\liminf \varphi_n(\omega) \geqslant 0\right\ ; \\ S_n(\omega)=X_1(\omega)+\cdots+X_n(\omega) \endgathered we have \lim _n \rightarrow \infty \fracS_n(\omega)-\left(p_1+\cdots+p_n\right)n=0 \quad \text a.e. in D .