You'd be better off drawing the graph with your raw (ungrouped) data. Grouping reduces detail, and therefore makes any correlation you find less reliable. If you've got lots of data pairs, you'll find software that does it for you easily enough. Chances are you've summarised your data into a small number of groups (<10?) in which case a scatter graph is of no value.
Correlation between discreet and continuous data isn't a problem, so long as you mention that your codomain for your regression formula has to be the integer set.
IQ & BMI eh?? So your hypothesis is that fat people are thick?? Here's hoping you find a low correlation coefficient, speaking personally!
Re 100 pairs in your data. The whole point of a scatter graph is to look for a trend. The more points you use, the more valid your observation (your regression line) will be. 100 points is ideal - enough to see a trend, not too many to be unmanageable. It takes time, but lengthy repetitive operations and calculations are an occupational hazard in statistics, which is why computers are used so widely.
PS Don't be scared to report the observation "no correlation". You may be looking for correlation, but if it isn't there, then reporting its absence is a result itself and a starting point for discussion.