It's a number written in hardware in scientific notation. Take 5.3*10^2. There's a mantissa (5.3) and an exponent (2) and a sign (+). This abstracts the precision (bits you give to the mantissa) from the range (bits you give to the exponent). For a double, you get around 15 digits of precision and a range up to 10^300ish. For floats, you get around 7 digits of precision and a range up to 10^40ish.
Floating point arithmetic is tricky to get right. It has many non intuitive limitations. Google around for tutorials.
Search for IEEE 754 if you're curious about how it works on a bit level. |