If your nominal IV only has two values then you should treat it as a scale variable and give them the values of 0 and 1. If you have more than two values then you will need to create several binary "dummy" variables like the one described above.
For example, if your nominal variable has three values a, b and c you will need two dummy variables: one for the value a and the other for the value b. An a is represented by 1, 0. A b is represented by 0, 1 and a c is represented by 0, 0. You always need one less dummy variable than the number of ordinal values.
Just to elaborate on Peter's point. You can conceptually think of dummy coding as providing a regression equation for each group.
e.g., Y = 2.2 + 1.2X1 + 3.4X2 + e
If X2 is dummy coded then you can represent this as:
Y = 2.2 + 1.2X1 for the group coded zero
Y = 5.6 + 1.2X1 for the group coded one
In fact any binary coding scheme will work (but the coefficient for X2 and the intercept will shift under different schemes) such as effect coding {-1,1} or {-0.5, 0.5}. However, even odd coding like 1 = male, 2 = female will resolve correctly into the same equation for each group.
Note that these lines will be parallel - to have non-parallel lines you can add additional terms such as X1*X2 interaction (sometimes called a test of parallelism or homogeneity of covariance in ANCOVA).
I think you can follow Peter Samuels and can practice more through you tube. Just remember you will get one less value in your final outcome if its >2 values.