There are a lot of short comments on this thread, so let me try to clear up any confusion.  Mann-Whitney would be an appropriate test for you to use.  However you should understand the hypothesis it tests.  If the two groups have similar shape and spread, it test for a difference in medians.  If this is not the case, it tests for stochastic dominance.  This is still related to the "location" of the data.  You had originally stated that you wanted to test for a difference in means.  You could still do this by permutation test (which may or may not be easy depending on what software you use).  However, if one of your samples is  highly skewed, looking at the mean may not be that meaningful anyway.

More fundamental comments:

Ultimately, you should ask yourself what hypothesis you really want to test.  We often just think about "testing for a difference", so we use t-test unless that's not appropriate, and then use Mann-Whitney, and not think much about it.  However, it is better to decide what hypothesis you actually want to test.  For example, among two samples, you could easily find that they have statistically different means, similar medians, similar 25th percentiles, and different 75th percentiles.  (We might expect these results if one sample is normal and one is right-skewed.)  Which of these hypotheses were actually of interest?  It depends on what you want to know. 

For example, looking at household income of two locations, we might expect the data to be skewed, and so not think means are very useful.  For a comparison of the two locations in general, the median income might suffice.  But for some government programs or, say, marketing campaign, we may want to look at the poorest segment (25th percentile) or wealthiest segment (75th percentile).

.

you might just use a bootstrap approach

under the null hypothesis, there is no gender difference ; you repeatedly draw with replacement a sample of size 56 in your population, randomly split it in two (complementary) subsamples of sizes 33/23 and compute the difference in mean between the subsamples

if you draw enough samples (say 1000), it gives you a good approximation of the sampling distribution of the difference in mean between a group of size 33 and its complement group (of size 23) taken from a sample of size 56

then you compare your empirical value to this sampling distribution and conclude

.

Or, a different approach

Article TO DETERMINE SKEWNESS, MEAN AND DEVIATION WITH A NEW APPROAC...

Depending on the software you use, a re-sampling based test, as Fabrice Clerot suggests, may be available in the package.  It might be called a permutation test.

You did not describe how the one sample was non-normal.  If it is strongly skewed, looking at the mean may not be very useful.

There are other tests that people use to compare two groups, for example by their medians.  Mann-Whitney is a common nonparametric test, but make sure you understand the hypothesis being tested.

Dear Samir

The statistical test of the null hypothesis of two independent samples and a non-normal distribution, you can use non-parametric test (Mann-Whitney), but in case, of the normal distribution, you can use two independent samples T-test.

Good Luck

Thank you Mr. Khalid. Since my data is non-normal, I am moving ahead with Mann-Whitney U test.

Thanks Mr. Fabrice Clerot, but my objective is to find difference gender-wose only.

Thank you Mr. Guven Mehmet Gunver  for sharing your article and Mr. Salvatore S. Mangiafico  for your input.

MW tests for the median only if both distributions have the same shape which is not the case

MW will test for stochasric dominance in your case which is not bad but definitely different

Mr. Fabrice, any other test that can be conducted when one sample is normal and the other is not?

Depending on your hypothesis a non-parametric test such as Wilcoxon sign rank test might be appropriate.

There are a lot of short comments on this thread, so let me try to clear up any confusion.  Mann-Whitney would be an appropriate test for you to use.  However you should understand the hypothesis it tests.  If the two groups have similar shape and spread, it test for a difference in medians.  If this is not the case, it tests for stochastic dominance.  This is still related to the "location" of the data.  You had originally stated that you wanted to test for a difference in means.  You could still do this by permutation test (which may or may not be easy depending on what software you use).  However, if one of your samples is  highly skewed, looking at the mean may not be that meaningful anyway.

More fundamental comments:

Ultimately, you should ask yourself what hypothesis you really want to test.  We often just think about "testing for a difference", so we use t-test unless that's not appropriate, and then use Mann-Whitney, and not think much about it.  However, it is better to decide what hypothesis you actually want to test.  For example, among two samples, you could easily find that they have statistically different means, similar medians, similar 25th percentiles, and different 75th percentiles.  (We might expect these results if one sample is normal and one is right-skewed.)  Which of these hypotheses were actually of interest?  It depends on what you want to know. 

For example, looking at household income of two locations, we might expect the data to be skewed, and so not think means are very useful.  For a comparison of the two locations in general, the median income might suffice.  But for some government programs or, say, marketing campaign, we may want to look at the poorest segment (25th percentile) or wealthiest segment (75th percentile).