A Java applet that clearly shows the difference between the distribution of the members of a dataset (dates of US pennies - skewed right) and the distribution of sample means.
http://statweb.calpoly.edu/chance/applets/SampleData/SampleData.html

This applet teaches students in a very dramatic way how changing the "bin width" of a histogram can dramatically alter the shape of the histogram. Lesson: you should graph your data with a number of different bin widths and choose the one that you feel best represents the data.

http://www.stat.sc.edu/~west/javahtml/Histogram.html

This is a clever and engaging version of Galton's Quincunx. It shows the underlying theoretical distribution (a normal curve) as well as the results of dropping mice into the quincunx.

http://www.mathcats.com/explore/bag-o/mice.html

Exploring Data is a website that I created in 1997 while working as the Raybould Fellow for Education Queensland.

It is a bit long in the tooth but the paper-based resources on the website are still useful. If you teach introductory statistics, it is worth a look around.
http://exploringdata.cqu.edu.au/index.htm

A clever web page for constructing a bar graph on the fly. All the necessary elements are editable - scales, axis labels, title, etc.
http://www.amblesideprimary.com/ambleweb/mentalmaths/grapher.html

This flash program shows the consistency of sample means from variously shaped populations. A student can change sample sizes, view raw data, choose samples singly or by the thousands, and observe changes as the sample numbers and populations change. It is also available as a stand-alone program (recommended for the classroom as it can be shown in full screen mode).

http://www.teacherlink.org/content/math/interactive/flash/sampling/sampl...

Line up the people in order of height. Then find the median, Q1 and Q3. For some reason they all look very cranky...
http://www.subtangent.com/maths/median-and-quartiles.php

A wonderful interactive version of this classical device.
http://www.subtangent.com/maths/ig-quincunx.php

The Stroop Test is a fun and engaging way to collect numerical data for later analysis and display. From the flipchart: "In the Stroop Test, the participants are given a list of words, and are asked to say the colour of the word, and not the word itself ... Most people take longer to say the colour of the word when the colour of the word doesn't match the word itself."

http://www.babynamewizard.com/namevoyager/lnv0105.html
The Baby Name Wizard is a fascinating website that graphically displays the number of babies with different names over the years. You can easily track the popularity of different names. What do you think was the most common boy's name and most common girl's name in 1990?

Note that the website takes some time to appear as it has to download a lot of data. Be patient!