Do you mean, how to exclude exactly 5% of cases having one or more missing values?
If so, then:
1. Include a variable (e.g., "status") to indicate whether each case has or doesn't have any missing data. (e.g., code as something like: 0 = no, 1 = yes)
2. Run the select cases process twice:
a. Selecting as live only the cases that have missing data, and saving to a new file.
b. Selecting exactly 95% of those cases, using random sampling.
The syntax would look like this:
*
* The following pulls only cases with missing data and saves them
* into a new file, "Missing_only"
*
*
DATASET COPY Missing_only.
DATASET ACTIVATE Missing_only.
FILTER OFF.
USE ALL.
SELECT IF (status = 1).
EXECUTE.
DATASET ACTIVATE DataSet2.
Now, open the newly created file ("Missing_only"), and use Data/Select cases/Random sample of cases/Exactly X of the first Y cases.
For Y, insert the total number of cases in the new file.
For X, insert whatever integer equals 95% of Y.
Then proceed however you wish (e.g., deleting unselected cases, or saving selected cases to a different file).