That depends on your input signal. For example, if you had a single, large glitch during a sample period, an average may not be a very accuracte representation of the signal. However, if the signal was relatively stable over the sample period, an average might be a good representation of the signal.
In peak detect mode, the scope oversamples at 250MS/s and saves the min and the max of each sample period. So for example, if the scope is sampling at 10MS/s and is in peak detect mode, it saves the min and the max of every 25 points that are oversampled at 250MS/s, or in other words the input is sampled every 4 ns, and the scope saves the min and max of the input over each 100 ns period. This mode is useful if you would like to be able to acquire for a long time, but still have a good chance of seeing glitches in your signal.
In sample mode, the scope saves every sample it acquires. If it's configured to sample at 250MS/s, it saves the state of the input every 4 ns. If it's configured to acquire at 10MS/s, if saves the state of the input every 100 ns. To get the most accurate representation of the waveform, use sample mode and acquire as fast as possible (zoom in to the smallest time per division in the app possible for your signal).
