This article reports the result of a research dealing with, 1) the difference between Puppet and Picture in teaching speaking, 2) the difference between students having high self-esteem and students having low self-esteem, and 3) the interaction between teaching media and self-esteem. The samples were two classes namely experimental class taught by using Puppet and control class taught by using Picture. Each class was divided into two groups in which each consisted of students having high self-esteem and those having low self-esteem. To collect the data, two instruments were used namely speaking test and self-esteem questionnaire. The data were, then, analysed by using descriptive statistics to describe the research data and inferential statistics namely multifactor Analysis of Variance ANOVA 2x2 and Tukey test to test the hypotheses. The findings of this research are: (1) Puppet is more effective than Picture to teach speaking; (2) The students having high self-esteem have better speaking skill than those having low self-esteem; and (3) There is an interaction between teaching media and students' self-esteem in teaching speaking. Based on the research findings, it could be summarized that Puppet is an effective media in teaching speaking to the eighth grade students of one of junior high schools in Semarang Regency in the academic year of 2014/2015. The effectiveness of the media is affected by self-esteem.