If you took twenty separate samples from a population and calculated their twenty slightly different means, you would find that these means are themselves normally distributed if you plotted them in a frequency graph. The standard error is somewhat like the standard deviation of these different estimates of the population mean.