Consider Bias as error on training data and Variance as error on test Data for different training samples.

Under fitting model:
High Bias and Low Variance [If you try to fit a simple model such that most of the training data points won’t be satisfied].

Over fitting model:
Low Bias and High Variance [If you try to fit a model such that most of the training data points would be exactly satisfied].

So, it is very important to build a Perfect Model such that it satisfies most of the training data points and gives better results for the test data [Low Bias and Low Variance].

For detailed explanation, download the following notes on "Bias & Variance" Tradeoff:
https://github.com/ruthvikraja/Bias-Variance.git