Character level language model is a class of model which focuses its study on how to predict the next character given by all previous text in a paragraph or essay. And the central point is to model the probability distribution of the next character given by all previous characters, and makes prediction based on this conditional probability distribution. In this paper, 2 models, n-grams and Recurrent Neural Network, are used to build the probability distribution on different text sources. And their performance of making predictions are evaluated using perplexity metric.