diff --git a/tests/kernels/test_flash_attn.py b/tests/kernels/test_flash_attn.py
index 71f61c19dd95..3a0723c17a69 100644
--- a/tests/kernels/test_flash_attn.py
+++ b/tests/kernels/test_flash_attn.py
@@ -8,7 +8,7 @@
 from vllm.utils import seed_everything
 
 NUM_HEADS = [(4, 4), (8, 2), (16, 2)]
-HEAD_SIZES = [128, 256]
+HEAD_SIZES = [4, 60, 80, 128, 256]
 BLOCK_SIZES = [16, 32]
 DTYPES = [torch.float16, torch.bfloat16]
 # one value large enough to test overflow in index calculation.
diff --git a/vllm/attention/backends/flash_attn.py b/vllm/attention/backends/flash_attn.py
index 22d07c0a4f68..2d7f5bb605e0 100644
--- a/vllm/attention/backends/flash_attn.py
+++ b/vllm/attention/backends/flash_attn.py
@@ -164,8 +164,8 @@ def _(
 class FlashAttentionBackend(AttentionBackend):
 
     @staticmethod
-    def get_supported_head_sizes() -> List[int]:
-        return [32, 64, 96, 128, 160, 192, 224, 256]
+    def get_max_supported_head_size() -> int:
+        return 256
 
     @staticmethod
     def get_name() -> str:
@@ -641,11 +641,11 @@ def __init__(
             raise ValueError(
                 "Sliding window is not supported in FlashAttention.")
 
-        support_head_sizes = FlashAttentionBackend.get_supported_head_sizes()
-        if head_size not in support_head_sizes:
+        max_head_size = FlashAttentionBackend.get_max_supported_head_size()
+        if head_size <= 0 or head_size > max_head_size:
             raise ValueError(
                 f"Head size {head_size} is not supported by FlashAttention. "
-                f"Supported head sizes are: {support_head_sizes}.")
+                f"Maximum supported head size is: {max_head_size}.")
 
     def forward(
         self,
diff --git a/vllm/attention/selector.py b/vllm/attention/selector.py
index 30aa7cb311af..94946a68a3d0 100644
--- a/vllm/attention/selector.py
+++ b/vllm/attention/selector.py
@@ -248,8 +248,8 @@ def which_attn_to_use(
             from vllm.attention.backends.flash_attn import (  # noqa: F401
                 FlashAttentionBackend)
 
-            supported_sizes = FlashAttentionBackend.get_supported_head_sizes()
-            if head_size not in supported_sizes:
+            maximum_size = FlashAttentionBackend.get_max_supported_head_size()
+            if head_size <= 0 or head_size > maximum_size:
                 logger.info(
                     "Cannot use FlashAttention-2 backend for head size %d.",
                     head_size)