With this kind of information you can use the Central Limit Theorem to assume that the grades are normally distributed. Then you can see that your score is about half a standard deviation above the average, and so you're at about the 70% percentile (link to the WolframAlpha calculation is provided). So you're likely 7/10. But if the standard deviation is artificially high because the people who failed failed really hard (as evidenced by the low end of 48), then you may likely be 8 or 9.

You can see the issues. A distribution of 10 people probably is not normal, and so higher order statistics like skew and kurtosis would be required to get a better estimate. In order to know the probability of the rank, you need some kind of distributional assumption to go from (mean,standard deviation) to probabilities, but for 10 students it's hard to see what that distribution could be.

http://www.wolframalpha.com/input/?i=integrate+e%5E(-1%2F2+x%5E2)%2Fsqrt(2+pi)+from+-infty+to+.5328

Dear Chris,

Thanks for your reply. Very helpful.

Actually, I'm considering sth more general based on this exam example.

Let's suppose the distribution behavior is not known.The only boundary conditions given are:

1) Number of samples; (Ex., 10~10,000)

2) The highest and lowest values;

(Actually, we can get  probabilities from above two conditions.

For example, there are 5 samples with integer values in an order:Ａ＞＝Ｂ＞＝Ｃ＞＝Ｄ＞＝Ｆ; the highest and lowest values are 3, 1; so there are １０ true events fit for ３＞＝Ｂ＞＝Ｃ＞＝Ｄ＞＝１; For the value ２, the probabilities for it to be at Ｂ is ３／/１０, to be at Ｃ is ３/１０, to be at Ｄ is ４/１０)

Though I got an answer in above example, I still don't know the general formula to do it. Let me give a more general question: For 1,000 samples listed in adding order: A