According to network models of memory, when information is stored in memory, it is not stored separately and by itself, but together with all the other aspects of the situation that you percieved. For example, if you listen to music, you do not simply memorize the tune, but also your mood, the causes for that mood (your girlfriend next to you), the smells and sights of the place, even what you did before and after listening to the music. Each of these informations is stored as a knot or node in the "semantic network" that is your memory, and with each time that things happen in conjunction, the connection between these nodes is strengthened a little bit.
Now, if you retrieve information from memory (e.g. you hear the music again or come to the concert hall and remember the place or tune), the node for that information gets activated and that activation spreads along the connections of that node to other nodes and activates them as well. So for example listening to that piece of music again will remind you of all the notable situations, but also of things that get activated through other activated nodes. For example remembering your girlfriend sitting next to you at that one time, will remind you also of how she broke up with you a few weeks later, and thus the music may make you sad, although when you listened to it the first time, it was a happy situation.
If you google for "semantic network" you will find comprehensive explanations, illustrations and examples.
As to the answer by Oriesok Vlassky explaining episodic memory: Both models integrate easily and do not contradict each other but rather complement each other well. So I would say the answer is his and mine combined.
