Department of Electrical Engineering Universitas Indonesia has developed an automatic essay grading system called Simple-O since 2007. Simple-O uses the Latent Semantic Analysis (LSA) method to compare two essays by extracting the essay into matrix. The previous development of Simple-O is the addition of Learning Vector Quantization (LVQ) which is a method of artificial neural network. This research will discuss and provide analysis related to the effect of adding word similarity function to the automatic essay grading system (Simple-O) to the accuracy of the system itself. The experiment will be conducted with five different scenarios by varying the number of keywords in the student’s answer essay to 100%, 80%, 60%, 40%, and 20% of the reference essay keywords. According to the result, there are scenarios that has decreased and increased in accuracy. The average accuracy of the Simple-O system after the addition of word similarity function has increased, though not significant. The average increase in accuracy after the addition of word similarity function is 5.4% from 90.9% to 96.3%.