The Nyquist sampling theorem (sometimes called the Nyquist-Shannon sampling theorem) says, if you have a signal that is bandlimited with bandwidth $B$, then if you sample it with a sampling period $T_s$ strictly less than $1/2B$, then the original signal can be perfectly reconstructed from the samples.
We call the minimum sampling rate for ideal reconstruction, $f_N = 2B$ ($f_N$ being in samples per second and $B$ in hertz), the Nyquist limit.
If you sample a signal with a sample rate greater than the Nyquist limit, it is (in principle) possible to perfectly reconstruct the original continuous-time signal.
If you sample a signal with a sample rate below the Nyquist limit, you can not perfectly reconstruct the signal due to aliasing.
So if you want to retain "complete" information about the signal you are sampling, you must sample above the Nyquist limit.