When you are building a model, you would not want the model to be too simple that it fails to fit well on training data. This means high error on training data(High Bias). This scenario is also referred to as Underfitting. Neither you would want your model to be too complex that it captures the noise in the data as well(High Variance). This scenario is also referred to as Overfitting. We need to find the middle way, to fit the model just well. This is nothing but "Bias Variance Trade Off".