Nowadays, huge amount of information is being produced and stored in electronic format using Amharic language. Developing effective and efficient Amharic information retrieval system is essential for searching and retrieving relevant document written in Amharic language. The ambiguity in Amharic language and the problem of synonymy and polysemy of terms have impacts on the performance of Amharic text retrieval system. Different researches have been conducted to enhance the performance of the Amharic information retrieval system. However, they didn’t address relevance feedback technique to improve retrieval performance. This research is initiated to examine, explore and experiment the effectiveness of relevance feedback technique for improving Amharic text retrieval performance using the capabilities of vector space feedback system. To test the performance of the prototype system, the total of 450 Amharic documents and 12 test queries are used. The documents are collected from different websites. The total number of unique terms generated after preprocessing and automatic indexing are 31,875. For searching and comparison between documents and query terms, vector space model is used. The matching is done using cosine similarity. The two types of relevance feedback techniques, explicit and pseudo feedback, are incorporated into the vector space model. For query modification, the two ways of query reformulation, i.e., query expansion and term reweighting techniques are used. For term reweighting techniques we used Rocchio algorithm and for query expansion, new related terms which share similar or related meaning selected from retrieved relevant documents are added to initial user query in order to achieve better retrieval effectiveness. The performance of the prototype system is evaluated before and after relevance feedback in three groups with nine methods. 11-point interpolated average precision and mean average precision (MAP) is used to evaluate the performance. Based on the experiment, the highest MAP was registered in explicit feedback top 25 new related terms selected from the relevant documents are added to initial query for query expansion. The value of the highest MAP was 0.652 (65.2%). The lowest MAP value was registered in standard vector space model and the value was 0.539 (53.9%). The experiments have shown that relevance feedback approach outperforms the standard vector space model by 11.3%. Different forms of Amharic writing system and the combination of two words was the challenge.