This study presents an approach to automatic summarization of Amharic newsrntexts by extracting sentences in a give n document. The objective o f this study isrnto investigate the application of machine learning technique (naive Bayesrnmethod) to automatic summarization of Amharic news items. The focus is onrnhow to use the naive Bayes classifier for automatic Amharic news textrnsummarization to extra ct sentences , i. e. on how to train the na'ive Bayes tornclassify sentences from Amharic news textsrnrnFirst each sentence is represented by a set of p redefined features (attributes)rn(i .e. location of a sentence in a document, title words occurring in the sentence,rnand cue words occurring in the sentence) that Edmondson (1969) found as arngood indicator in giving an optimum summary for scientific papers. In addition,rnthe thematic words occurring in the sentence. Then the naive Bayes algorithm isrnused to train to classify sentences as "a summary" and "not - a summary" basedrnon the feature vectors.rnrnFor the purpose of this study 480 Amharic news articles is used . Evaluation ofrnthe result s of the experiments is done using 10-fold cross validation. Result ofrnthe experiment shows that the location feature gives the best result in thernclassification n o f sentences when using individual features. The results of differentrncombinations of feature sets in which location feature is included shows betterrnresults than when location is not included.rnrnBased on the feature values estimated on the training program for therncombination of all the features a prototype summarizer is developed whichrnextracts sentences to a desired compression level.