A box of 100 crayons has 50 blue, 30 red and 20 yellow. A handful of 10 that represents the box should have about 5 blue, 3 red and 2 yellow. If your handful is 9 blue and 1 red, it still came from the box, but it describes the box badly: someone looking only at your handful would think yellow crayons do not exist.
For people, the groups that matter are the ones that might answer the question differently. For a survey about moving the school start time, grade level matters, since older students may prefer to start later, and so does whether students ride the bus. A representative sample has each of those groups in roughly the population's proportions. If a school is 25% sixth graders, a representative sample of 80 students includes about 0.25×80=20 sixth graders.
Representative is not the same as large. A survey with 5,000 responses, 90% from one grade, represents that grade, not the school. It is not the same as random, either. Random selection is the method that usually produces a representative sample, but any single random sample can still come out lopsided, so you check the result's proportions against what you know about the population.
Think of it like this. A representative sample is like a museum's scale model of a city. Parks, roads and buildings take up the same share of space as in the real city, so you can learn the layout from it. A model missing a whole neighbourhood misleads however carefully the rest was built. Unlike a model, a sample only has to match on the features that matter for the question.
Core idea. A sample is representative when its key characteristics match the population's in the right proportions.