In [5]:

d.loc[(d['A'].isnull()) & (d.B == 't3'), 'A']='new_val'

d

Out[5]:

         A   B
0      NaN  t1
1       t2  t2
2  new_val  t3
3       t3  t4
4  new_val  t3

[5 rows x 2 columns]