Textual entailment can provide a semantic inference to natural language expression tasks that have meaning variability and it solves language ambiguity problems. Textual entailment classification task is to predict the given pairs of natural language expressions such that a human who reads the first element of a pair would most likely infer that the other element is entailment, neutral, or contradiction. rnIn this work, we developed a textual entailment classification by using deep learning approach. Amharic is one of the most under resourced language, and has challenges in its linguistic levels. We developed an end to end solution as the approaches using deep Neural networks that can eliminate the underlying extensive feature engineering that shown traditional approaches for the development of Amharic Textual entailment classifier. rnThe reverse side sentence matching model for Amharic textual entailment classification has five main layers. Which are word embedding, sentence embedding, and sentence matching, aggregation and prediction layers. Specifically, for the first word embedding layer, applied word vectors by attaching its sub word information that can represent the important features of Amharic words. Second In sentence embedding layer by leveraging bi-directional long short term memory network we are able to remember long sentence sequences of the dataset we prepared (pair of 8700 sentences) for Amharic Inference task. Third in matching layer we matched each time step of the hypothesis sentence against all time step of the premise by applying matching functions followed by a matching approaches, Our model produced encouraging result in contrast to the base line models by scoring 77% training and 72% of test accuracy,