This article describes a research project into the self-efficacy and anxiety of college English students at four universities in China. A total of 738 participants completed a questionnaire measuring self-efficacy and anxiety in writing in English. This was immediately followed by a writing task. The questionnaire used a seven point Likert type scale to measure self-efficacy and anxiety in writing. The questionnaire also included open ended questions concerning student perceptions of effort, actual effort and parental pressure. The quantitative data relating to self-efficacy and anxiety were analysed using structural modelling techniques. In the first instance, confirmatory factor analysis provided evidence for the validity of constructs. Subsequently a full structural model was hypothesised and tested. The hypothesized model indicated that both anxiety and self-efficacy predicted writing performance. However, in a re-specified model a better fit was achieved. The final model indicated that the relationship between writing performance and anxiety was mediated by self-efficacy. This supports Bandura (1986) social cognitive theory of learning that perceptions of affect can influence self-efficacy beliefs. From the open ended data the results indicated that anxious students were more likely to experience parental pressure, have low effort perceptions and low actual effort; those students with high efficacy were more likely to have high effort perceptions, were less likely to experience parental pressure and were likely to spend longer studying English. (Contains 5 figures and 11 tables.)