Researchers should consider not only student’s cognition load but their active cognitive processing when designing example in example learning. Researches on fading example and correct - incorrect paired example showed that both of them can improve student’s learning in near transfer but can’t let every student involving in active processing. Because of those shortcomings, some researchers pointed out that in order to initiating student’s active processing in example learning, explanations should be given to student with example at the same time. But contradictory conclusions were gotten with comparative studies on example learning between with and without teaching explanation, those between with and without self-explanation, those between with teaching explanation and self-explanation. Because contradictory conclusions were gotten from researches on different explanation in example learning, because fading example is chosen in most of example learning studies, and it increases student’s more cognition load, in our research, we chose correct - incorrect paired example in order to compare the efficiency of correct example and correct-incorrect paired example, compare the efficiency of self-explanation, instruction explanation, no explanation. We chose probability problems in mathematics in Grade 9 and designed learning materials of experiment 1 and experiment 2, instant post-test and delayed post-test. In experiment 1, between subject experiment is designed to compare the influence of correct example and correct - incorrect paired example to probability problem solving. 90 students were assigned randomly to correct example group, correct-incorrect example group and control group. The results show that students’ test performance in correct - incorrect paired example group are significantly higher than students’ in correct example and control group. There are no significant difference of students’ performance in correct example group and control group. In experiment 2, between subject experiment is designed to compare the influence of different explanation (instruction explanation, self-explanation, no explanation) to probability problem solving in correct - incorrect example learning. 90 students were assigned randomly to self-explanation group, instruction explanation group and no explanation group. The results show that students performance in self-explanation group and instruction explanation group are significantly higher than students in no explanation group in instant post-test and there are no significant difference of students’ performance between instruction explanation group and self-explanation group. But students’ performance in self-explanation group are significantly higher than students in instruction explanation group and no explanation group in delayed post-test, and there are no significant difference of students’ performance between instruction explanation group and no explanation group in delayed post-test. The conclusions are (1) correct-incorrect paired example is significantly more effective than correct example in statistics problem solving; (2) correct - incorrect paired example with explanation is significantly more effective than correct - incorrect paired example without explanation; (3) correct - incorrect paired example with self-explanation is significant better and lasted more time than correct - incorrect paired example with instruction explanation.