You can use one of the heteroskedasticity consistent methods for estimating standard error.

For a more detailed discussion, see Article Using Heteroskedasticity-Consistent Standard Error Estimator...

Of note, SPSS provides these methods in the GLM procedure (among these methods, HC3 or HC4 may be preferable; see the above-mentioned article).

Good luck.

It's really not helpful to use significance tests to test for things like normality or homoscedasticity as a pre-requisite for other tests. You are much better off using plots to assess if the assumptions of the analysis are reasonable. .... That being said, Ali Zia-Tohidi provided one possible approach for the problem. But if the heteroscedasticity is extreme, you might re-assess what approach makes sense.

To second Salvatore S. Mangiafico answer: the visual inspection may be helpful to indentify the cause for heterogeneity. A linear model may not be suitable to describe the data generating process, but instead some form of generalized linear model, which allow for other distributions of the errors.