Non-normal univariate distribution -- how to define "low" to "high" subsets of values? Sorry for the dumb question: it's been a million years since I took basic stats.
I have a set of 48, non-normally distributed values ranging from 1 to 100.  I'd like to be able to apply a (valid) descriptive scale to these values, grouping values into sets along the lines of these are "low", these are "below avg", "avg" "above avg" & "high". Is this possible with a non-normal distribution (and a small value set like this -- this set is the full population I'm interested in)? 
I was thinking in terms of quartiles and using the median rather than the mean -- but then would I have to stick to "low" "medium" and "high", using <Q1 as my "low" and>Q3 as my "high"? 
Given that this is essentially a qualitative, descriptive objective, how wrong would it be to call values < Q1 as "low", Q1 to Q2 as "below avg", Q2 to mean as "average", mean to Q3 as "above average" and> Q3 as "high"?
[If it helps, Q1=22.4, Q2=30.8, Q3=47.2, Q4=100; mean is 35.3; mode=33.33 ... so a lot of my values do centre in the median-mean range there, it's just that I have a long tail toward 100 and it does not make sense to dump the values.]
The full set of values is:
0.0 0.0 10.5    11.1    11.5    13.3    14.3    16.7    17.1    20.0    20.0    22.2    22.5    23.1    23.1    23.5    23.7    24.7    25.0    26.7    27.1    29.5    30.0    30.4    31.3    33.0    33.3    33.3    33.3    34.8    35.0    38.3    41.2    42.9    46.2    46.5    49.2    50.0    50.0    52.8    55.6    55.6    58.1    60.0    66.7    80.0    100.0   100.0
 A: You should be reassured that "keep it simple" is the advice just as much as it is your inclination. 
What follows may be more detailed or complicated than you want or need, but if it's along the right lines you can simplify too. 
All the 48 points are shown below at right. The rising scale corresponds to increasing cumulative probability and serves to shake them apart. 
The red reference line shows the mean. 
A box plot is shown at left, with median and quartiles (conventionally) and 5% and 95% points (less conventionally) defining the whiskers. 
The numbers shown as text (with some rounding to 1 d.p.) give the numeric story: 
The minimum and maximum are 0 and 100 and so forth. 
You can make qualitative comments too: there are two values a little detached at 0 and two values similarly detached at 100. The value at 80 is also notably high. Perhaps you can add substantive flesh. It's a matter of taste and judgment and experience whether to be dramatic and call them outliers. That judgment should depend on how surprising or exceptional they seem to you. For example, if these were exam scores as percents, two people were perfect and two flunked completely, and look at that!  
Meta-prejudices: 


*

*Dividing into bins arbitrarily based on mean or median, etc., at best imposes a framework arbitrarily. Feel free to use the mean, median, etc., as summary, but let the data show themselves. 

*Using cluster analysis (K-means etc.) on data like this is unduly technical here. 

A: In this case, quartiles or quintiles would be just fine. They're easy to understand and they automatically give you roughly equal group sizes.
There are also a number of more complicated approaches you could take. This paper studies a few of them and name-drops a whole bunch more. Of all the algorithms mentioned there, "k means" (and possibly its cousins "k medians" and "k medoids" or "partitioning around medoids") is the most widely used, most readily available, and easiest to understand and explain.
A: I would use either standard deviation or quartiles. If the distribution is not skewed, then whatever is less/more than one standard deviation would be low/high, and those within would be average/typical.
If the distribution is skewed, then Q1 is low, Q2-Q3 are average and Q4 is high.
